Error Propagation, Variance-Covariance and Error Ellipses - Memory Anchors
Memory techniques transform abstract formulas into vivid, retrievable images. Instead of re-reading derivations, your brain retrieves a story, a picture, or a punchy phrase in seconds. For board exam performance under time pressure, this is your secret weapon. The anchors below convert every key formula and concept in Error Propagation into something you simply cannot forget — even at 2 AM during the exam week.
Memory anchors
Law of Propagation of Variances — errors add in quadrature (squares), NOT linearly
Memory aid
Think of carrying two separate baskets of bibingka, each with a slight tilt (error). When you carry BOTH, the combined wobble is NOT twice as bad — it's less, because the tilts don't always go the same direction. Mathematically, the combined error is √(σ₁² + σ₂²), not σ₁ + σ₂. Two imperfect bakers don't make disaster twice as bad — they make it √2 times as bad.
Why it works
The familiar Filipino food analogy creates a sensory, everyday image. The wobble metaphor captures the vector-like nature of random errors that partially cancel.
Example usage
When the board exam asks for the error of a sum of two independently measured distances (σ₁ = 0.03 m, σ₂ = 0.03 m), recall the wobbling baskets: combine under the square root → √(0.03² + 0.03²) = 0.0424 m, NOT 0.06 m.
Recall trigger
Wobbling bibingka baskets
Error propagation formula: σ_y² = Σ(∂f/∂xᵢ)² σᵢ²
Memory aid
Remember the phrase: 'PARTIAL SQUARED SIGMA SQUARED' — PSS². Every term in error propagation has TWO squarings: the partial derivative is SQUARED, and the sigma is SQUARED. Say it like a chant: P-S-S-SQUARED, P-S-S-SQUARED. The result is also a SQUARED quantity (variance), so squares are everywhere.
Why it works
Repetition of 'squared' reinforces the most common board exam mistake — forgetting to square the partial derivative. The chant rhythm encodes it in procedural memory.
Example usage
For A = s², ∂A/∂s = 2s. Apply PSS²: σ_A² = (2s)² · σ_s². With s = 50 m, σ_s = 0.02 m: σ_A² = (100)² × (0.02)² = 4.0 m⁴ → σ_A = 2.0 m².
Recall trigger
PSS² chant
Scaling rule: if y = kx, then σ_y = |k|σ_x
Memory aid
A jeepney driver tells you 'my GPS error is ±5 m.' If you triple the fare zone distance (k = 3), the billing error also triples to ±15 m. Scaling your measurement scales your uncertainty exactly the same way. The constant k 'stretches' or 'compresses' the error band proportionally.
Why it works
The jeepney fare zone is a Filipino context students instantly picture. The linear stretch concept is intuitive and directly mirrors the formula.
Example usage
If a distance of x = 200 m has σ_x = 0.05 m, and you compute y = 3x, then σ_y = 3 × 0.05 = 0.15 m.
Recall trigger
Jeepney fare zone × 3
Sum of n equal observations: σ_sum = σ√n (not σ/√n)
Memory aid
Mang Kiko tapes a 60-m traverse in 6 equal 10-m segments, each measured with σ = 0.01 m. He asks his student: 'Ano ang error ng buong traverse?' The student blurts 'σ/√6!' Mang Kiko sighs: 'Hindi! That is the error of the MEAN. For the SUM — when you walk forward and errors pile up — it is σ√n = 0.01√6 = 0.0245 m.' The SUM grows (√n multiplies), the MEAN shrinks (√n divides). Walk FORWARD = errors pile up = multiply. Walk BACK to average = errors shrink = divide.
Why it works
The micro-story presents the most common board exam trap (confusing sum vs. mean) through a relatable mentor-student dialogue. The physical metaphor 'walk forward' vs. 'walk back to average' differentiates the two cases.
Example usage
Board question: A traverse leg is taped in 5 equal segments each with σ = 0.02 m. Find σ of the total. Answer: σ√5 = 0.02 × 2.236 = 0.0447 m (NOT 0.02/√5).
Recall trigger
Mang Kiko's traverse tape segments
Mean of n observations: σ_mean = σ/√n
Memory aid
To find the MEAN, divide by √n — your error gets lean! To find the SUM, multiply by √n — your error gets big again! Mean = lean (shrinks). Sum = drum (beats louder, grows).
Why it works
Rhyme encodes two opposing rules simultaneously with a memorable contrast. 'Lean' and 'drum' create auditory and visual cues that differentiate the two formulas.
Example usage
4 readings of a distance, each σ = 0.02 m. σ_mean = 0.02/√4 = 0.01 m. If summed instead, σ_sum = 0.02√4 = 0.04 m.
Recall trigger
Mean = lean, Sum = drum
Matrix form of error propagation: Σ_y = J Σ_x J^T
Memory aid
Visualize a SANDWICH: J is the top bread slice, Σ_x is the filling, and J^T is the bottom bread slice. The Jacobian wraps around the covariance matrix like a pandesal around the filling. The sandwich is the propagated covariance. You cannot eat the filling alone (raw covariance) — you need the Jacobian bread on both sides to propagate it through a function.
Why it works
The pandesal/sandwich image is culturally resonant and encodes the correct operator order (J on the left, J^T on the right) through a spatial left-right placement metaphor.
Example usage
When correlated observations exist, write: Σ_y = J · Σ_x · J^T. The Jacobian J contains all ∂f/∂xᵢ partial derivatives as a row vector (for a scalar function).
Recall trigger
Pandesal sandwich with Σ_x filling
Variance-covariance matrix: diagonal = variances, off-diagonal = covariances
Memory aid
Picture a Class Record matrix. The diagonal (your own row-column pairing) shows YOUR OWN grades (variances — how uncertain you are about yourself). The off-diagonal shows how YOUR grade correlates with your classmate's grade (covariance — how your error relates to another parameter's error). A zero off-diagonal means no correlation — your classmates' grades are independent of yours.
Why it works
Filipino students are very familiar with class records. The self-vs-classmate framing perfectly captures the diagonal (self) vs. off-diagonal (relationship with others) structure.
Example usage
After least squares adjustment, the covariance matrix of [X, Y] coordinates has σ_X² and σ_Y² on the diagonal, and σ_XY (covariance) off the diagonal. If σ_XY ≠ 0, the X and Y errors are correlated.
Recall trigger
Class record matrix — diagonal = self, off-diagonal = classmate correlation
Adjusted covariance matrix formula: Σ_x̂ = σ̂₀² (A^T P A)^{-1}
Memory aid
Remember the acronym 'PAPA-INVERSed' → P·A^T·P·A inverted: the matrix (A^T P A) is your 'PAPA' (father/PAPA in Filipino) — strong and central to the family. You invert PAPA and multiply by the estimated variance factor σ̂₀² to get the covariance of your unknowns. If PAPA is well-conditioned (strong geometry), the inverted PAPA is small (small uncertainties). Weak geometry = weak PAPA = huge inverted PAPA = large uncertainties.
Why it works
'PAPA' is a strongly emotional Filipino word, making it highly memorable. The analogy of a strong/weak father directly maps to well/poorly conditioned normal equations.
Example usage
After forming the normal equations N = A^T P A, the covariance of adjusted unknowns is Σ_x̂ = σ̂₀² · N⁻¹. The diagonal elements are the variances of each adjusted unknown.
Recall trigger
Inverting PAPA (A^T P A)
Error ellipse — 2D position uncertainty visualized as an ellipse
Memory aid
Imagine throwing darts at a target (survey point). If you have equal skill in both X and Y directions, the darts form a CIRCLE of uncertainty. But if you always miss more in one direction (weak geometry along one axis), the darts form a STRETCHED ELLIPSE. The error ellipse is just the dartboard pattern of where your adjusted position could land.
Why it works
Dart-throwing is universally understood. The circle-vs-ellipse contrast directly embeds the concept of isotropic vs. anisotropic uncertainty. It also naturally explains why elongated = weak geometry.
Example usage
When interpreting GNSS results over PRS92, a nearly circular error ellipse means the satellite geometry (PDOP) is good in all directions. A highly elongated ellipse (say, N-S stretched) means the north-south position is much less certain than east-west.
Recall trigger
Darts on a target — circle or stretched ellipse
Error ellipse axes come from eigenvalues of the 2×2 covariance matrix
Memory aid
Professor Lim draws two perpendicular arrows on the board and says: 'The covariance matrix is like a spinning top. Its EIGENVALUES tell you how fat it is in each principal direction — the bigger eigenvalue is the LONG axis of the ellipse, the smaller eigenvalue is the SHORT axis. Find the eigenvalues, take their square roots — those are your semi-axes.' A student asks, 'What's the orientation angle?' Professor Lim winks: 'The eigenvectors — they point where the ellipse leans.'
Why it works
The classroom micro-story reinforces the exam-critical link between eigenvalues→semi-axis lengths and eigenvectors→orientation angle through a conversational Q&A format.
Example usage
Given Σ = [[4, 1],[1, 2]] (units: mm²), find eigenvalues λ₁ and λ₂. Semi-axes of the standard error ellipse: a = √λ₁, b = √λ₂.
Recall trigger
Professor Lim's spinning top — eigenvalues = axis lengths
Standard error ellipse corresponds to ~39% probability (NOT 95%)
Memory aid
The standard error ellipse is 'the BARELY-PASSING ELLIPSE' — 39% is like almost failing (passing is 40%!). It is NOT 95%. To get 95% confidence, you must SCALE UP by a factor of about 2.45. Remember: 'Standard = 39% = almost failed; scale by 2.45 to PASS with 95% confidence.'
Why it works
Philippine board exam passing is 70%, but students viscerally understand '39% = failing.' The emotional association with a failing grade makes 39% immediately memorable and distinct from the intuitive but wrong answer of 68% or 95%.
Example usage
A board question states: the semi-axes of the covariance ellipse are computed as a = 5 mm, b = 3 mm. What are the 95% confidence ellipse axes? Answer: a₉₅ = 2.45 × 5 = 12.25 mm, b₉₅ = 2.45 × 3 = 7.35 mm.
Recall trigger
39% = almost failed; ×2.45 = 95% confidence
Elongated error ellipse = weak geometry in that direction
Memory aid
Picture a taho vendor's trolley wheel that is very narrow (thin ellipse) in one direction. The trolley wobbles badly when pushed sideways (weak in that direction) but is stable fore-and-aft. An elongated error ellipse is exactly like that wobbly wheel — it shows the direction where the survey geometry cannot hold steady.
Why it works
The taho vendor is a vivid Filipino street scene. The directional stability metaphor directly maps to the geometric interpretation of ellipse elongation.
Example usage
In a GNSS survey using PRS92/WGS84, if the error ellipse is elongated in the north-south direction, it means there are fewer or weaker satellite signals in that azimuth range — the north-south component of position is least reliable.
Recall trigger
Wobbly taho trolley wheel — elongated = wobbly direction
Partial derivative is the 'sensitivity factor' — how much y changes per unit change in xᵢ
Memory aid
Think of a cellphone signal booster. The partial derivative ∂f/∂xᵢ is the GAIN of the booster — if your input signal has noise (σ_xᵢ), the booster amplifies BOTH the signal AND the noise by the same gain factor. A high partial derivative (high gain) means a small input error causes a LARGE output error. Always check your partial derivatives — a gain of 100 turns a 0.01-unit error into a 1.0-unit error!
Why it works
The signal booster metaphor is intuitive for Filipino students who are smartphone-savvy. The gain concept directly mirrors the role of ∂f/∂xᵢ as an amplifier of uncertainty.
Example usage
Area A = L × W. ∂A/∂L = W = 40 m (gain of 40). If σ_L = 0.03 m, this contributes (40 × 0.03)² = 1.44 m⁴ to σ_A².
Recall trigger
Cellphone signal booster gain = partial derivative
Jacobian matrix J — matrix of all partial derivatives
Memory aid
The Jacobian is the 'Report Card of Sensitivities.' Each row is one output function, each column is one input variable. The entry in row i, column j is: 'How sensitive is output i to a change in input j?' Visualize a school report card grid — subjects on one axis, students on the other. The Jacobian is your sensitivity report card.
Why it works
Report card grids are instantly visualizable for Filipino students. The row-column structure of the report card maps directly to the row-function, column-variable structure of the Jacobian.
Example usage
For two functions f₁ = x₁ + 2x₂ and f₂ = 3x₁ − x₂, the Jacobian is J = [[1, 2],[3, −1]]. This is then used in Σ_y = J Σ_x J^T.
Recall trigger
Sensitivity report card — Jacobian
Covariance σ_xy — positive means errors tend to go in the same direction; negative means opposite
Memory aid
Two surveyors, Jobert and Noel, always take readings together. When Jobert reads high (positive error), Noel also tends to read high — positive covariance. But if Jobert reads high and Noel reads low (like they're on opposite ends of a seesaw), the covariance is negative. Zero covariance = they read independently, with no seesaw relationship.
Why it works
The seesaw image is universally understood and physically demonstrates negative covariance. The paired-surveyor scenario makes abstract statistical correlation concrete.
Example usage
In a traverse adjustment using PPCS/UTM, if the Easting and Northing adjustments are positively correlated (σ_EN > 0), an overestimate in E tends to accompany an overestimate in N — this is captured in the off-diagonal of the covariance matrix.
Recall trigger
Jobert-Noel seesaw — same direction = positive, opposite = negative covariance
Unit weight variance σ̂₀² = V^T P V / (n − u) — the a posteriori variance factor
Memory aid
Remember 'VPOV over DOF': V^T P V divided by degrees of freedom (n − u). Think: 'Vee-POV over DOF.' VPV is the weighted sum of squared residuals — your 'view' (POV) of how badly the observations fit the model. Divide by DOF to normalize per degree of freedom.
Why it works
The acronym 'Vee-POV' is quirky and memorable. 'POV' is a modern term Filipino students use daily (point of view), making a technical formula feel relatable.
Example usage
If n = 10 observations, u = 6 unknowns, and V^T P V = 0.0016, then σ̂₀² = 0.0016 / (10 − 6) = 0.0004, so σ̂₀ = 0.02.
Recall trigger
Vee-POV over DOF
Independent observations — covariance matrix is diagonal (no off-diagonal terms)
Memory aid
Imagine 5 CCTV cameras positioned at different PPCS/UTM control points. Each camera records independently — no camera 'talks' to another. Their error matrix is DIAGONAL — each camera has its own variance, but no cross-terms. If you networked them (correlated readings), off-diagonal terms appear. Independent observations = cameras that don't network = diagonal covariance matrix.
Why it works
CCTV cameras are ubiquitous in Philippine cities. The networked vs. standalone metaphor elegantly captures why independence produces a diagonal matrix.
Example usage
Uncorrelated angle measurements in a triangulation network have Σ_L = σ² I (identity-scaled), so the weight matrix P = σ⁻² I is also diagonal.
Recall trigger
Standalone CCTV cameras = diagonal covariance matrix
Common board pitfall: adding errors linearly instead of in quadrature
Memory aid
When errors combine, don't just add in a line — PUT THEM UNDER THE SQUARE ROOT SIGN! σ₁ + σ₂ is WRONG, √(σ₁² + σ₂²) is STRONG! Say this every time you see two independent errors combining: 'Under the root, that's the route!'
Why it works
The rhyme encodes the rule AND the common mistake simultaneously. The 'route' rhyme with 'root' creates an auditory hook that fires during recall.
Example usage
Two survey legs with σ₁ = 0.04 m and σ₂ = 0.03 m: Wrong answer = 0.04 + 0.03 = 0.07 m. Correct answer = √(0.04² + 0.03²) = √(0.0025) = 0.05 m.
Recall trigger
'Under the root, that's the route!'
Error ellipse orientation angle θ — direction of the major axis
Memory aid
The error ellipse is like a tilted buko (coconut) lying on a table. The orientation angle θ is how much the coconut is tilted from North (or the X-axis). The formula involves the off-diagonal covariance σ_xy: when σ_xy = 0 (diagonal matrix), the coconut stands upright (ellipse axes align with coordinate axes). When σ_xy ≠ 0, the coconut tilts. tan(2θ) = 2σ_xy / (σ_x² − σ_y²) — the tilt formula.
Why it works
The buko coconut is culturally resonant. The tilt metaphor physically captures what changing θ means, and the upright vs. tilted cases directly correspond to σ_xy = 0 vs. σ_xy ≠ 0.
Example usage
If σ_x² = 9 mm², σ_y² = 4 mm², σ_xy = 2 mm²: tan(2θ) = 2(2)/(9 − 4) = 4/5 = 0.8, so 2θ = 38.66°, θ = 19.3° from the X-axis.
Recall trigger
Tilted buko coconut — orientation angle of error ellipse
Propagating error through area of a rectangle: σ_A² = (W·σ_L)² + (L·σ_W)²
Memory aid
A cadastral surveyor under RA 4374 (old) and now RA 8560 (new) measures a lot: Length L = 80 m (σ = 0.03 m) and Width W = 40 m (σ = 0.02 m). 'How uncertain is the lot area?' The formula: A = L × W, ∂A/∂L = W = 40, ∂A/∂W = L = 80. So σ_A² = (40 × 0.03)² + (80 × 0.02)² = 1.44 + 2.56 = 4.00 m⁴. σ_A = 2.0 m². 'Kailangan nating mag-ingat sa width measurement — it contributes MORE to area uncertainty!'
Why it works
The cadastral context under RA 4374/RA 8560 grounds the problem in Philippine professional practice. The final insight (width contributes more) trains students to interpret which term dominates, a frequent board question.
Example usage
Board question: A lot is 80 m × 40 m. σ_L = 0.03 m, σ_W = 0.02 m. Find σ_A. Use PSS²: σ_A = √[(40×0.03)² + (80×0.02)²] = √[1.44 + 2.56] = √4.00 = 2.0 m².
Recall trigger
RA 8560 cadastral lot — rectangle area error
Revision game
I am the rule that says you CANNOT just add two errors together — you must square both, add, then take the root. What rule am I?
Answer
The Law of Propagation of Variances (quadrature addition rule)
Memory link
A1 — wobbling bibingka baskets; A18 — 'Under the root, that's the route!'
I am a matrix where the diagonal tells you how uncertain you are about YOURSELF, and the off-diagonals tell you how your errors relate to your neighbor. What matrix am I?
Answer
The Variance-Covariance Matrix
Memory link
A7 — Class Record Matrix analogy
I measure the 2D position uncertainty as an egg-shaped figure. I have a major axis, a minor axis, and an orientation angle. What am I?
Answer
The Error Ellipse
Memory link
A9 — dart-throwing target; A10 — Professor Lim's spinning top
I am the percentage probability covered by the STANDARD error ellipse — and I am LESS THAN HALF. What percentage am I?
Answer
39%
Memory link
A11 — 'Standard = 39% = almost failed'
You tape a 50-m traverse in 10 equal segments, each with σ = 0.01 m. I am the standard deviation of the TOTAL length. Am I σ√10 or σ/√10?
Answer
σ√10 = 0.01 × √10 = 0.0316 m (the SUM grows — use σ√n)
Memory link
A4 — Mang Kiko's traverse segments; A5 — Mean=lean, Sum=drum
I am the matrix sandwich formula — J on the left, Σ_x in the middle, and J-transpose on the right. Write me.
Answer
Σ_y = J Σ_x J^T
Memory link
A6 — Pandesal sandwich with Σ_x filling
A board exam asks: the error of the MEAN of 9 equal observations each with σ = 0.03 m. What is the answer, and what is the most common wrong answer?
Answer
Correct: σ/√9 = 0.03/3 = 0.01 m. Common wrong answer: 0.03√9 = 0.09 m (that is the SUM, not the MEAN!)
Memory link
A4, A5 — Mang Kiko's story; Mean=lean rhyme
I connect the adjusted unknowns' covariance to the design matrix, weight matrix, and a posteriori variance. I am written as σ̂₀² × (something)⁻¹. What am I?
Answer
Σ_x̂ = σ̂₀² (A^T P A)^{-1}
Memory link
A8 — Inverting PAPA (A^T P A)
Formula mnemonics
σ_y² = (∂f/∂x₁)²σ₁² + (∂f/∂x₂)²σ₂² + ···
Mnemonic
PSS² — Partial Squared, Sigma Squared. Every term has TWO squarings. Chant: 'Partial-Squared, Sigma-Squared, sum them all up!'
When to use
Whenever you compute a quantity y from independently measured quantities x₁, x₂, … and need to find the standard deviation or variance of y.
What each part means
σ_y² = variance of the derived quantity y; (∂f/∂xᵢ)² = squared partial derivative (sensitivity factor); σᵢ² = variance of the i-th independent observation.
σ_sum = √(σ₁² + σ₂²) for y = x₁ ± x₂
Mnemonic
'Under the root, that's the route!' Both sigmas go UNDER THE SQUARE ROOT. Never add them directly.
When to use
When two independently measured quantities are added or subtracted — e.g., two traverse legs added to get total length, or two angles summed.
What each part means
σ_sum = standard deviation of the sum or difference; σ₁, σ₂ = standard deviations of the two independent observations.
σ_y = |k| σ_x for y = kx
Mnemonic
'Scaling multiplies — the constant RIDES the sigma.' Whatever constant k scales your measurement, k also scales your error. Think: jeepney fare ×3 = error ×3.
When to use
When computing a derived quantity that is a simple scalar multiple of a measured quantity — e.g., converting units, computing a scaled distance.
What each part means
k = constant multiplier; σ_x = standard deviation of measurement; |k| ensures we always get a positive result.
σ_sum = σ√n (n equal, independent observations summed)
Mnemonic
'SUM = SIGMA TIMES ROOT N — errors GROW when you SUM.' Contrast with the mean: 'MEAN = SIGMA OVER ROOT N — errors SHRINK when you AVERAGE.'
When to use
For the SUM: when n equal-precision measurements are chained/summed (e.g., taped segments in a traverse). For the MEAN: when n measurements of the same quantity are averaged.
What each part means
σ = standard deviation of each individual observation (assumed equal); n = number of observations; σ√n grows with n; σ/√n shrinks with n.
Σ_y = J Σ_x J^T
Mnemonic
'PANDESAL SANDWICH — J is bread, Σ_x is filling, J^T is the other bread slice.' Always put J on the LEFT and J^T on the RIGHT.
When to use
When propagating errors through a system of functions with potentially correlated input variables — the general matrix form of error propagation.
What each part means
Σ_y = covariance matrix of output quantities; J = Jacobian matrix of partial derivatives; Σ_x = covariance matrix of input quantities; J^T = transpose of J.
Σ_x̂ = σ̂₀² (A^T P A)^{-1}
Mnemonic
'Sigma-hat-squared times PAPA-INVERSE.' PAPA = A^T P A. Invert PAPA, multiply by σ̂₀², get the covariance of your adjusted unknowns.
When to use
After least squares adjustment to quantify the uncertainty and correlation of adjusted coordinates or parameters.
What each part means
Σ_x̂ = covariance matrix of adjusted unknowns; σ̂₀² = a posteriori variance of unit weight; A = design matrix; P = weight matrix; (A^T P A)^{-1} = inverse of the normal equation matrix.
σ̂₀² = V^T P V / (n − u)
Mnemonic
'Vee-POV over DOF.' VPV = weighted sum of squared residuals = your 'point of view' of how well the model fits. Divide by degrees of freedom (n − u) to get the unit weight variance.
When to use
To compute the reference variance after least squares adjustment — needed to scale the covariance matrix of adjusted unknowns.
What each part means
V^T P V = weighted sum of squared residuals; n = number of observations; u = number of unknowns; (n − u) = redundancy / degrees of freedom.
tan(2θ) = 2σ_xy / (σ_x² − σ_y²) — orientation of error ellipse
Mnemonic
'Twice the angle, twice the covariance, divided by the variance difference.' Think: TAN-2θ = 2-COV over VAR-DIFF. The factor of 2 appears everywhere — two times angle, two times covariance.
When to use
When finding the orientation of the error ellipse from a 2×2 covariance matrix of position coordinates.
What each part means
θ = orientation angle of the major axis from the X-axis; σ_xy = covariance; σ_x², σ_y² = variances of X and Y; the formula gives the principal axis direction.
Quick recall chains
Steps to Propagate Error Through a Function
Recall test
Without looking: What are the 6 steps to propagate error through y = f(x₁, x₂)? Trigger: LECHON!
Memory chain
Remember 'WRITE, DERIVE, SQUARE, MULTIPLY, SUM, ROOT' — like building a LECHON: Write the recipe, Derive the ingredients (partial derivatives), Square the meat portions, Multiply by seasoning (σᵢ²), Sum in the pot, and Root (roast) it to perfection (σ_y).
Items to remember
- 1. Write the function y = f(x₁, x₂, …)
- 2. Compute all partial derivatives ∂f/∂xᵢ
- 3. Square each partial derivative
- 4. Multiply each squared partial by the corresponding σᵢ²
- 5. Sum all terms to get σ_y²
- 6. Take the square root to get σ_y
Error Ellipse Parameters to Find
Recall test
What are the 6 steps to fully define an error ellipse from a covariance matrix? Trigger: EROS shooting arrows!
Memory chain
Remember 'EXTRACT, EIGEN, ROOT, ROOT, ORIENT, SCALE' — EEROS! Like EROS the love god shooting arrows: Extract the covariance (pull back the bow), find Eigenvalues (aim), Root them for axes (release), Root again (second arrow), Orient the ellipse (love has direction!), and Scale to 95% (maximum range).
Items to remember
- 1. Extract the 2×2 covariance submatrix [σ_x², σ_xy; σ_xy, σ_y²]
- 2. Compute eigenvalues λ₁ ≥ λ₂
- 3. Semi-major axis a = √λ₁
- 4. Semi-minor axis b = √λ₂
- 5. Orientation angle θ: tan(2θ) = 2σ_xy/(σ_x² − σ_y²)
- 6. Scale for confidence: ×2.45 for 95%
Key Distinctions: Sum vs. Mean Error
Recall test
A traverse has 9 equal segments, each σ = 0.01 m. What is σ of the total length? What is σ of the mean segment? Trigger: Snowball vs. spreading rice!
Memory chain
Think 'GROWING SUM, SHRINKING MEAN, STRETCHED SCALE, QUADRATURE PAIR.' SUM is like a snowball rolling downhill (grows). MEAN is like spreading rice on a tray (thinner = smaller error). SCALED is like a zoomed photo (same ratio). QUADRATURE PAIR is like two perpendicular legs of a triangle (Pythagorean combination).
Items to remember
- Sum of n equal observations: σ_sum = σ√n (grows)
- Mean of n equal observations: σ_mean = σ/√n (shrinks)
- Scaled quantity y = kx: σ_y = |k|σ_x
- Sum of two unequal observations: σ = √(σ₁² + σ₂²)
Board Exam Pitfall Checklist for Error Propagation
Recall test
Name the 5 most common pitfalls in error propagation board exam problems. Trigger: LaPiS na Error Sa Sürveying!
Memory chain
Remember 'LINEAR, PARTIAL, SUM-MEAN, ELLIPSE-PERCENT, SIGMA-HAT' → Acronym: L-P-S-E-S = 'LaPiS na Error Sa Sürveying' (stone of error in surveying). Every pitfall is a stone that trips you on the exam!
Items to remember
- Pitfall 1: Adding errors linearly instead of in quadrature
- Pitfall 2: Forgetting to square the partial derivative
- Pitfall 3: Using σ/√n (mean) instead of σ√n (sum)
- Pitfall 4: Treating the standard ellipse as 95% probability (it's 39%)
- Pitfall 5: Forgetting to multiply by σ̂₀² when computing covariance matrix of adjusted unknowns
Matrix Form Error Propagation: Σ_y = J Σ_x J^T — Order of Operations
Recall test
Write the Jacobian for f = x₁² + 3x₂ with inputs x₁ and x₂, then propagate the error using matrix form. Trigger: Ibis landing on the branch!
Memory chain
IDENTIFY-IDENTIFY-BUILD-SET-LEFT-RIGHT-RESULT → 'IIBSeLReR' → 'Ibis (bird) LANDS Left then Right on the branch' — the Ibis first lands from the left (J · Σ_x) then settles to the right (· J^T). A pandesal sandwich assembled left-bread first, then right-bread.
Items to remember
- 1. Identify all output functions f₁, f₂, …
- 2. Identify all input variables x₁, x₂, …
- 3. Build Jacobian J: rows = functions, columns = variables
- 4. Set up Σ_x (input covariance matrix)
- 5. Compute J · Σ_x (left multiply)
- 6. Multiply result by J^T (right multiply)
- 7. Result Σ_y is the output covariance matrix
Diagrams
Error Propagation and Variance-Covariance — Complete Concept Map
A comprehensive mind map showing all major topics in this chapter: the Law of Propagation (scalar and matrix forms, special cases), the Variance-Covariance Matrix (structure and origin from least squares), the Error Ellipse (parameters, confidence levels, geometric meaning), and common board exam pitfalls. Use this as a navigation map before and after studying.
Step-by-Step Error Propagation Decision Flowchart
A decision-tree flowchart guiding students through the correct error propagation method based on the type of problem. Starting from identifying the derived quantity, the flowchart branches through: single vs. multiple inputs, scaled vs. non-scaled, independent vs. correlated, and sum vs. mean cases. This is the universal solver for board exam problems.
Error Ellipse Construction and Interpretation Flowchart
A step-by-step flowchart for constructing an error ellipse from a 2×2 covariance matrix. The process flows from extracting the covariance submatrix, computing eigenvalues, finding semi-axes and orientation angle, selecting confidence level, and interpreting the ellipse shape geometrically. Particularly relevant for GNSS and geodetic network adjustments under PRS92/WGS84.
Confidence Level Distribution for Error Ellipses
A pie chart showing the probability coverage of different error ellipse confidence levels. The dominant visual takeaway is that the standard error ellipse (39%) does NOT cover the majority of probability — students must always clarify which confidence level is required and apply the appropriate scale factor.
Error Propagation Knowledge Build-Up: From Observations to Error Ellipse
A timeline showing the logical progression from raw field observations to the final error ellipse in a geodetic adjustment. This connects all chapter concepts in a linear workflow, helping students see how error propagation, variance-covariance matrices, and error ellipses are part of one integrated process — not isolated topics.