Least Squares — Observation Equations - Memory Anchors
For abstract mathematical concepts like least squares adjustment, vivid analogies, clever mnemonics, and memorable micro-stories create neural 'hooks' that make retrieval fast and reliable — exactly what you need in a timed PRC board examination. The anchors below transform dry matrix algebra into unforgettable mental images, stories, and word tricks. Work through each anchor actively: visualize it, say it aloud, and test yourself with the recall trigger before moving on.
Memory anchors
Least squares minimizes the weighted sum of squared residuals (∑wv²)
Memory aid
Imagine you are a basketball coach and each player's missed free throw is a 'residual.' A clumsy big man (low weight) missing by 2 meters matters less than your star shooter (high weight) missing by 2 centimeters. You square the misses so big blunders get punished extra hard, then you minimize the total pain. That is least squares — punish blunders, respect precision, and minimize the total weighted suffering.
Why it works
The sports analogy connects abstract minimization to a concrete, emotionally resonant scenario. Squaring = extra punishment for big misses. Weights = importance of each player/observation. Filipino students who follow the PBA or UAAP will immediately relate.
Example usage
Board exam asks: 'What does least squares minimize?' Think of the basketball coach → minimize weighted squared misses → ∑wv².
Recall trigger
Think: 'Star shooter vs. clumsy big man = weighted residuals'
Residual v = computed minus observed (v = Ax̂ − l)
Memory aid
Remember 'CO': Computed minus Observed = residual. Think of CO as 'Company Owes' — the company (computed value) owes something relative to what was actually observed. The 'debt' is the residual v.
Why it works
The acronym CO is short, the phrase 'Company Owes' gives it a story, and the sign convention (computed − observed) is baked into the imagery.
Example usage
When setting up residual equations, always write v = Ax̂ − l (computed − observed), never flip the sign. Trigger: 'CO, computed first.'
Recall trigger
CO → Computed minus Observed → v
The design matrix A contains partial derivatives of each observation function with respect to each unknown
Memory aid
The design matrix A is like a Philippine government org chart (organigram). Each row is one barangay (observation). Each column is one department (unknown parameter). The cell entry tells you how sensitive that barangay is to that department's actions — that is a partial derivative. If a barangay has zero in a column, that department does not affect it at all.
Why it works
Org charts are familiar to Filipino students from civics and public administration contexts. The row = observation, column = unknown mapping is made concrete.
Example usage
Setting up a leveling network: each level route is one row; each benchmark correction is one column; the entry is +1 or −1 depending on direction.
Recall trigger
Org chart → rows are observations, columns are unknowns, entries are sensitivities (partials)
Normal equations: (AᵀPA)x̂ = AᵀPl
Memory aid
Chant the formula as: 'ATPA times x̂ equals ATPl' — say it like a rap beat: 'A-T-P-A, solve for x, A-T-P-L, that's the text!' The pattern AᵀP_A on the left and AᵀP_l on the right is symmetric: both start with AᵀP, but one ends in A and the other in l.
Why it works
Rhythmic chunking (rap-style repetition) leverages phonological memory loops. The parallel structure (both sides start with AᵀP) makes it self-checking.
Example usage
On the board exam, when asked to write the normal equations, start with AᵀP on both sides — left gets A appended, right gets l appended.
Recall trigger
Rap beat: 'A-T-P-A … A-T-P-L'
Solution: x̂ = (AᵀPA)⁻¹ AᵀPl
Memory aid
Remember the solution as 'INVERT, then APPLY': Invert the normal equation matrix (AᵀPA)⁻¹, then apply it to the right-hand side AᵀPl. The phrase: 'INVERT the boss, APPLY to the workers' — (AᵀPA) is the boss (normal matrix), AᵀPl is the workers (data-driven vector).
Why it works
Two-step action verbs (INVERT, APPLY) are easier to sequence than pure algebra. The boss/worker metaphor encodes the matrix relationship.
Example usage
After forming normal equations, the answer is always: invert the left-side matrix, multiply by the right-side vector. That is x̂.
Recall trigger
INVERT the boss → APPLY to workers → x̂
Redundancy (degrees of freedom) = n − u (observations minus unknowns)
Memory aid
Story: Engineer Maria has n = 10 GNSS baseline observations to solve u = 6 coordinate unknowns in a PRS92 network. She has 4 'leftover' observations she did not strictly need. Those 4 extras are her redundancy — her safety net. Without them, she cannot check her work. 'n minus u gives you your safety net' — always n − u.
Why it works
A relatable Filipino engineering scenario with a named character makes the formula personal and memorable. The 'safety net' metaphor encodes why redundancy matters.
Example usage
12 observations, 8 unknowns → redundancy = 12 − 8 = 4. You need at least 1 redundancy to adjust.
Recall trigger
Maria's safety net: n minus u = redundancy
Reference variance: σ̂₀² = vᵀPv / (n − u)
Memory aid
Think of σ̂₀² as the teacher's grade curve. After the exam (adjustment), the teacher checks: 'Did I make the test too hard or too easy?' vᵀPv is the total weighted pain (sum of weighted squared residuals). Dividing by n − u (degrees of freedom) gives the average pain per extra observation. If the result ≈ 1, the test was fair (weights were correct). If large, the test was too hard (or there's a blunder).
Why it works
Filipino students are very familiar with grade curves and teacher assessments. The metaphor maps perfectly: total error / degrees of freedom = fairness check.
Example usage
After adjustment, compute σ̂₀² = vᵀPv/(n−u). If ≈ 1.0, weighting was realistic. If >> 1, suspect blunders.
Recall trigger
Teacher's grade curve: vᵀPv divided by (n − u)
Single unknown = weighted mean: x̂ = Σwᵢlᵢ / Σwᵢ
Memory aid
Picture a traditional Filipino weighing scale (timbangan) with multiple fruit vendors placing their price quotes (observations) on pans. The heavier the vendor's reputation (weight wᵢ), the more their quote tilts the balance. The equilibrium point of the scale is the weighted mean — the least-squares answer for one unknown.
Why it works
The timbangan is a culturally familiar object in Philippine markets. The physical equilibrium metaphor directly encodes the weighted mean formula.
Example usage
Three level routes give elevations 152.34, 152.30, 152.36 m with weights 0.5, 1.0, 0.25. Compute: (0.5×152.34 + 1.0×152.30 + 0.25×152.36)/(0.5+1.0+0.25) = 152.32 m.
Recall trigger
Timbangan (weighing scale) in the palengke → weighted mean
Weight is inversely proportional to variance: w = 1/σ² (or w ∝ 1/K for leveling routes)
Memory aid
A precise GPS receiver (low variance σ²) is a trustworthy witness in court — you give it HIGH weight. A rusty barometer (high variance) is an unreliable witness — LOW weight. Weight is the inverse of unreliability. The less variance, the more you trust it, the higher the weight: w = 1/σ².
Why it works
The court-witness analogy captures the concept of trustworthiness inversely scaling with uncertainty. Filipino legal dramas make this relatable.
Example usage
Level route of 2 km vs. 8 km: w₁ = 1/2 = 0.5, w₂ = 1/8 = 0.125. Shorter route = more weight.
Recall trigger
Trustworthy witness = high weight; unreliable witness = low weight → w = 1/σ²
Need n > u to perform least-squares adjustment (redundancy must be positive)
Memory aid
Story: A judge (least squares) can only render a verdict if there is more evidence (observations n) than suspects (unknowns u). With exactly as many suspects as clues (n = u), you get a unique but unverifiable answer — no error checking possible! You need at least ONE extra clue (n > u) to catch inconsistencies. Board exam trap: n = u means no adjustment, just a unique solution.
Why it works
The court-case narrative is engaging and culturally familiar. The 'extra clue = redundancy' link is logically tight and memorable.
Example usage
If a problem has 6 observations and 6 unknowns, state: 'No adjustment possible; unique solution, zero redundancy.'
Recall trigger
Judge needs more clues than suspects: n > u
Weight matrix P is diagonal for independent (uncorrelated) observations
Memory aid
Picture a Filipino chess board. Only the diagonal squares are lit up — everything off-diagonal is dark. That is the weight matrix P for independent observations: a diagonal blaze of weights, zero everywhere else. If observations are correlated, some off-diagonal squares light up too.
Why it works
The chess-board image is iconic and spatially encodes the mathematical structure of a diagonal matrix in a memorable way.
Example usage
When observations are independent, build P = diag(w₁, w₂, …, wₙ). No cross terms.
Recall trigger
Chess board diagonal only lit → P is diagonal for independent obs
Linearization: observation equations are linearized using Taylor series (first-order partials)
Memory aid
Linearization is like approximating a hilly mountain trail (curved function) with a series of short flat boardwalks (tangent lines). Each boardwalk is only valid near its starting point — your approximate values x₀. The partial derivatives in matrix A are the slopes of those boardwalks. Far from x₀, the approximation breaks down — you need iteration.
Why it works
The mountain-trail-to-boardwalk image makes the mathematical concept of Taylor linearization tangible and visually vivid.
Example usage
In a trilateration network, observed distances are nonlinear functions of coordinates. Linearize around approximate coords x₀, iterate until corrections are negligible.
Recall trigger
Mountain trail → flat boardwalks → linearization via partials
The vector l = observed minus computed at approximations (l = ℓ − f(x₀))
Memory aid
l is the 'Leftovers' vector — what is left over after you subtract what your approximate model predicts from what you actually measured. 'l = Leftovers = Observed minus Computed-at-x₀.' Note: l and v have opposite sign: l is o−c, v is c−o.
Why it works
The 'leftovers' metaphor is universally relatable (Filipino culture values using all food — leftovers are the remainder). The sign contrast with v is highlighted.
Example usage
If a measured distance is 250.342 m and the distance computed from approximate coordinates is 250.315 m, then l = 250.342 − 250.315 = 0.027 m.
Recall trigger
l = Leftovers: observed minus computed at approximations
Unweighted case (all weights equal) → simple arithmetic mean
Memory aid
When all the weights are set to one, The arithmetic mean gets done! No heavy, no light — all equal play, The mean of the data shows the way. But give some more trust, give others less — Then weighted mean handles the rest!
Why it works
Rhyming couplets exploit phonological memory. The rhyme encodes both the equal-weight case (arithmetic mean) and the unequal-weight case (weighted mean) as a contrast pair.
Example usage
If all observations have equal precision (same σ), set all wᵢ = 1. Then x̂ = Σlᵢ/n = arithmetic mean.
Recall trigger
Rhyme: 'All weights equal one → arithmetic mean is done'
Checking σ̂₀² ≈ 1 validates the weight system
Memory aid
Story: Quality Control Inspector Noel is checking a bridge bolt factory. After production (adjustment), he measures a sample and computes the average deviation. If it matches the machine's claimed precision (σ̂₀² ≈ 1), the machine is calibrated correctly. If the deviation is way off, the machine lied about its precision — just as bad weights in a least-squares network produce σ̂₀² >> 1, signaling the weight model is wrong or there is a blunder.
Why it works
Quality control is an engineering concept Filipino students encounter in multiple subjects. The factory inspector character makes the abstract statistical test concrete.
Example usage
After adjustment, if σ̂₀² = 4.7, conclude that weights were too optimistic (precisions overestimated) or a blunder exists — re-examine the network.
Recall trigger
Inspector Noel's QC check: σ̂₀² ≈ 1 means calibrated correctly
Observation equation form: v = Ax̂ − l (the fundamental parametric model)
Memory aid
Walk through the Geodetic Engineering building at your university. At the ENTRANCE: see a 'v' (residual) written on the door — you always start with residuals. Inside Room 1 (A): the DESIGN MATRIX room — full of partial derivatives. In Room 2 (x̂): the UNKNOWNS room — the adjusted parameters. At the EXIT (l): the LEFTOVERS bin — observed minus computed. The equation is the path: v = A·x̂ − l.
Why it works
Method of loci (memory palace) encodes algebraic structure into spatial memory, which is processed by a different brain system and is highly persistent.
Example usage
During an exam, mentally 'walk the building' to reconstruct v = Ax̂ − l before setting up any least-squares problem.
Recall trigger
Walk the GE building: entrance v → Room A → Room x̂ → exit l
Leveling route weight ∝ 1/K where K = route length in km
Memory aid
A longer commute from Quezon City to Taguig (long route) is more tiring and error-prone than a short trip within Makati (short route). The shorter the commute (K), the more reliable the level route, the higher the weight. Weight = 1/commute distance. Short trip → light pack → high trust → high weight.
Why it works
Metro Manila commuting is a daily reality for many Filipino students. The length-of-commute metaphor maps perfectly onto route-length weighting.
Example usage
Level route A: K = 2 km → wA = 0.5. Level route B: K = 4 km → wB = 0.25. Route A gets more weight.
Recall trigger
Short Makati trip = high weight; long QC-to-Taguig = low weight → w = 1/K
AᵀPA is the normal equation matrix (also called the coefficient matrix of normal equations)
Memory aid
Chunk it as three layers: A-transpose wraps around P in the middle, then A again on the right. Like a sandwich: A (top bread) | P (filling) | A (bottom bread) = AᵀPA. The sandwich is square and symmetric — always! It must be invertible for a unique solution.
Why it works
Food chunking (sandwich) is a universal mnemonic. The symmetric sandwich structure also encodes the mathematical property that AᵀPA is symmetric positive definite.
Example usage
When computing normal equations, assemble the AᵀPA sandwich first, then compute its inverse. A 2×2 sandwich for 2 unknowns.
Recall trigger
Sandwich: A-top | P-filling | A-bottom = AᵀPA
Iteration is needed when the observation functions are nonlinear
Memory aid
Story: Surveyor Ben is locating a buried BM using GPS in a dense forest near the NAMRIA office. His first coordinate estimate is rough. He runs the least-squares adjustment, gets corrections Δx, updates x₀ ← x₀ + Δx, then runs it again. Each cycle, the corrections shrink. When corrections are smaller than 0.001 m (sub-millimeter), he stops — convergence achieved. Iteration = refining the approximation until the boardwalks match the mountain.
Why it works
A named character in a recognizable Philippine setting (NAMRIA context) makes the iterative process vivid. The convergence criterion (small corrections) is explicitly included.
Example usage
For nonlinear geodetic networks (trilateration, GPS baseline adjustment), always iterate: update approximations, recompute A and l, solve, repeat until corrections converge.
Recall trigger
Ben updates x₀ each cycle until Δx < tolerance → iteration
The parametric (observation equation) method vs. the condition equation method
Memory aid
Parametric method (observation equations): You write equations FROM THE OBSERVATIONS, solving for the UNKNOWNS directly. It is like filling out a BIR tax form — you enter your known data and compute your tax liability (unknowns). Condition method: You write equations that the ADJUSTED OBSERVATIONS must satisfy, using correlates. It is like writing the rules of a budgeting game before you play — constraints first, then find compatible values. Board exam: the parametric method is more common and easier to set up.
Why it works
The BIR tax form is a universally recognized Filipino experience. The contrast between the two methods is encoded as two distinct familiar activities.
Example usage
Most board exam problems use the parametric/observation equation method. Set up v = Ax̂ − l for each observation.
Recall trigger
BIR form = parametric (observations → unknowns); budgeting rules = condition method
Revision game
I am the most probable value when you have ONE unknown measured many times. I balance heavy and light measurements like a timbangan in the palengke. What formula am I?
Answer
The weighted mean: x̂ = Σwᵢlᵢ / Σwᵢ
Memory link
A8 — Timbangan (weighing scale) analogy
I am the vector of the 'leftovers' — the difference between what you actually observed and what your approximate model predicted. My sign is observed minus computed. Name me.
Answer
The l vector (observed minus computed at approximate values)
Memory link
A13 — 'Leftovers' mnemonic
A geodetic engineer has 15 GPS baselines and 10 coordinate unknowns. How many redundant observations does she have, and can she perform a least-squares adjustment?
Answer
Redundancy = 15 − 10 = 5; yes, adjustment is possible since n > u
Memory link
A6 — Maria's safety net story; formula r = n − u
I am the sandwich matrix — the 'boss' that must be inverted to solve for the unknown corrections. I am square, symmetric, and positive definite. What is my formula?
Answer
N = AᵀPA (the normal matrix)
Memory link
A18 — Sandwich chunking mnemonic
After adjustment, you compute me and find I equal 8.3. This is much larger than 1, so you panic. What am I telling you, and what should you do?
Answer
σ̂₀² = 8.3 >> 1 indicates blunders or over-optimistic weights; re-examine the data and weight model
Memory link
A15 — Inspector Noel's QC check story
A level route of 2 km gives an elevation of 34.500 m, and a level route of 8 km gives 34.520 m. Without computing, which observation gets MORE weight and why?
Answer
The 2 km route gets more weight (w = 1/2 = 0.5 vs. w = 1/8 = 0.125) because shorter routes accumulate less error — w = 1/K
Memory link
A17 — Metro Manila commuting analogy
I am the matrix that organizes how sensitive each observation is to each unknown parameter. My rows correspond to observations and my columns to unknowns. I am built from partial derivatives. What am I?
Answer
The design matrix A
Memory link
A3 — Government org chart analogy
You have n = 6 observations and u = 6 unknowns. Your classmate says you should run a least-squares adjustment. Are they correct? What is the fundamental rule?
Answer
No — with n = u, redundancy = 0 and no adjustment is possible; only a unique (un-checkable) solution exists. Need n > u for adjustment.
Memory link
A10 — Judge needs more clues than suspects story
Formula mnemonics
v = Ax̂ − l
Mnemonic
VAL: V equals A-hat minus L. 'VAL always starts the party' — Residual (V), design matrix times unknowns (A-x̂), minus leftovers (L). Or: 'Verkada Ang Lahat' — V = A·x̂ − l.
When to use
This is the fundamental observation equation. Write one row per observation when setting up any least-squares problem in parametric form.
What each part means
v = vector of residuals (computed minus observed); A = design matrix (partials of observations w.r.t. unknowns); x̂ = vector of adjusted parameter corrections; l = vector of observed minus computed at approximate values (the 'leftovers').
(AᵀPA)x̂ = AᵀPl
Mnemonic
The ATPA ATPL chant: 'A-T-P-A solves for x, A-T-P-L is the text.' Both sides start with AᵀP — left gets A, right gets l. Remember: left is the coefficient, right is the constant.
When to use
After assembling A, P, and l, multiply out to form normal equations. Solve for x̂ using matrix inversion.
What each part means
Aᵀ = transpose of design matrix; P = diagonal weight matrix (inverse variances); A = design matrix; x̂ = unknown corrections; l = observed-minus-computed vector. AᵀPA is the normal matrix (square, symmetric). AᵀPl is the right-hand side vector.
x̂ = (AᵀPA)⁻¹ AᵀPl
Mnemonic
'INVERT THE BOSS, APPLY TO WORKERS': (AᵀPA)⁻¹ = invert the boss; AᵀPl = the workers (data). Combine: x̂ = boss⁻¹ × workers.
When to use
Use after forming normal equations. For a 2×2 or 3×3 system, invert manually. For larger systems, use Gaussian elimination or a calculator.
What each part means
(AᵀPA)⁻¹ = inverse of the normal matrix (the 'coefficient of unknowns'); AᵀPl = transformed observation vector; x̂ = least-squares adjusted parameter corrections.
σ̂₀² = vᵀPv / (n − u)
Mnemonic
'VPV over freedom': V-P-V (residuals weighted and summed) divided by degrees of freedom (n minus u). Think: 'VPV / DF = fairness score.' The numerator vᵀPv is the weighted sum of squared residuals (total punishment); the denominator n − u is how many redundant checks you had.
When to use
Always compute after adjustment to check the quality of the solution. Compare to expected value (should be ≈ 1 for a properly weighted, blunder-free network).
What each part means
v = vector of final residuals; P = weight matrix; vᵀPv = weighted sum of squared residuals (a scalar); n = number of observations; u = number of unknowns; n−u = degrees of freedom (redundancy).
x̂ = Σwᵢlᵢ / Σwᵢ (weighted mean, single unknown)
Mnemonic
'Timbangan Formula': Sum of (weight × reading) on top, sum of weights on bottom. It is literally how you balance a scale — heavier weight pulls the average toward its reading. 'W-L over W' — weights times levels, divided by weights.
When to use
Only when there is ONE unknown observed multiple times (e.g., multiple level routes to one benchmark, repeated angle measurements at one station). This is the most common board-exam application.
What each part means
wᵢ = weight of i-th observation; lᵢ = i-th observed value; Σwᵢlᵢ = sum of weighted observations; Σwᵢ = sum of all weights; x̂ = most probable (least-squares) value.
w ∝ 1/K (leveling), w = 1/σ² (general)
Mnemonic
'Short and Light vs. Long and Heavy': Short route (small K) → small denominator → large weight (light, easy to trust). Long route (large K) → large denominator → small weight (heavy, hard to trust). General: variance (σ²) is the denominator — less variance = more weight.
When to use
Assign weights before computing weighted mean or setting up P matrix. For leveling networks, use w = c/K where c is a constant (often set to make one route have weight 1).
What each part means
K = route length in km (for differential leveling); σ² = variance of the observation; w = weight assigned to the observation. For leveling, accumulated error grows with distance, so longer routes get less trust.
r = n − u (redundancy / degrees of freedom)
Mnemonic
'n minus u = your safety net r': n observations MINUS u unknowns = r redundant checks. No net if n = u. Big net if n >> u. 'Safety Net Size = n minus u'.
When to use
Before any adjustment: count n and u. If n ≤ u, stop — you cannot adjust. Always state the redundancy in your solution for full marks.
What each part means
n = total number of independent observations; u = number of unknown parameters; r = redundancy (degrees of freedom). If r = 0: unique solution, no checking. If r < 0: under-determined, no solution.
Quick recall chains
Steps to Solve a Least-Squares Problem (Parametric Method)
Recall test
Without looking, list the 10 steps of the parametric least-squares procedure. Can you name what you compute at step 8 and 9?
Memory chain
Story chain: 'Identify Maria (1-2), Linearize her Path (3-4), Pack her Weights (5), Form the Normal Sandwich (6), Invert the Boss (7), Find the Residual Leftovers (8), Check the QC Score (9), Update and Repeat (10).' Each bold word is an action step.
Items to remember
- 1. Identify unknowns (u) and observations (n); confirm n > u
- 2. Choose approximate values x₀ for the unknowns
- 3. Linearize each observation equation; compute partials → build matrix A
- 4. Compute l = observed − computed at x₀
- 5. Assign weights; build diagonal P matrix
- 6. Form normal equations: N = AᵀPA, t = AᵀPl
- 7. Solve: x̂ = N⁻¹ t
- 8. Compute residuals: v = Ax̂ − l
- 9. Compute reference variance: σ̂₀² = vᵀPv/(n−u)
- 10. Update x₀ ← x₀ + x̂; iterate if nonlinear
Board-Exam Traps in Least Squares (Common Pitfall Sequence)
Recall test
Name 4 common board-exam pitfalls in least-squares adjustment from memory. What does n = u imply?
Memory chain
Acronym: 'SPONWS' — Sign (v), Sign (l opposite), P diagonal, nequals u no-go, One unknown = mean, Watch σ̂₀². Remember: 'SPONWS — Sponge the Errors Away from the Board Exam!'
Items to remember
- Sign of v: computed minus observed (not reversed)
- Sign of l: observed minus computed (opposite of v)
- P is diagonal only if observations are independent
- n = u means NO adjustment (zero redundancy)
- Weighted mean only works for ONE unknown
- σ̂₀² ≈ 1 is the check, not σ̂₀² = 0
Conditions for a Valid Least-Squares Adjustment
Recall test
State 5 conditions required for a valid least-squares adjustment. What happens if AᵀPA is singular?
Memory chain
Remember 'MORE-NICE-RANDOM-FULL-INVERT': More obs than unknowns → Independent obs → Random normal errors → Full rank A → Invertible AᵀPA. 'MORE NICE RANDOM FULL INVERT = a perfect least-squares setup!'
Items to remember
- n > u (more observations than unknowns)
- Observations must be independent (or correlation accounted for in P)
- Errors must be random and normally distributed
- A must have full column rank (rank = u)
- AᵀPA must be invertible (non-singular)
Types of Weights Used in Philippine Geodetic Practice
Recall test
How do you assign weights to (a) leveling routes, (b) repeated angle measurements, and (c) GNSS baselines?
Memory chain
Story: 'LRAGE weights: Level routes by Length, Repeated measurements by Number, Angles by Variance, GNSS by Covariance, Equal weights give Average.' L-R-A-G-E = 'LRAGE — Large Surveys Need Good Effort to Weight Properly!'
Items to remember
- Leveling routes: w = 1/K (K = km)
- Repeated measurements: w = n (number of repetitions)
- Angle observations: w = 1/σ² (σ from set statistics)
- GNSS baselines: w = 1/σ² from covariance matrix
- Equal precision: w = 1 for all (arithmetic mean)
The Four Key Matrices/Vectors in Observation Equations
Recall test
Name all 5 key matrices/vectors. State the dimension of each in terms of n (observations) and u (unknowns).
Memory chain
Acronym: 'APLXv' — A-P-L-X-v. Chant: 'All Pupils Learn X Variables' where X = x̂ and V (residuals) = the verdict. Dimensions: A is n-by-u, P is n-by-n, l and v are n-by-1, x̂ is u-by-1.
Items to remember
- A — Design matrix (partials, n×u)
- P — Weight matrix (diagonal, n×n)
- l — Observed minus computed vector (n×1)
- x̂ — Unknown corrections vector (u×1)
- v — Residual vector (n×1)
Diagrams
Least Squares Observation Equations — Complete Concept Map
A hierarchical mind map showing all major concepts of the least-squares observation equation method: the governing principle, the five key matrices/vectors, the observation equation, normal equations, special cases, quality checks, and when iteration is needed.
Step-by-Step Procedure: Parametric Least-Squares Adjustment
A detailed flowchart of the full parametric least-squares adjustment procedure. Shows decision points for checking redundancy (n > u), evaluating reference variance quality (σ̂₀² ≈ 1), and determining whether iteration is needed for nonlinear models. Includes feedback loops for blunder detection and iterative refinement.
How to Assign Weights in Philippine Geodetic Practice
A decision-tree flowchart for assigning weights to different types of geodetic observations common in Philippine practice: leveling routes (w = 1/K), repeated measurements, GNSS baselines, and equal-precision observations. All paths lead to assembling the diagonal P matrix.
State Diagram: Least-Squares Adjustment Lifecycle
A state diagram showing the complete lifecycle of a least-squares adjustment: from initial setup through redundancy check, linearization, normal equations, solution, residual computation, quality check, and final convergence. Shows the feedback states for blunder detection and iterative refinement.