Discrete updates, cycles and boundaries
VectorPaint two-state Markov chain: row probabilities approach but do not reach stationarity in finite steps
Check a quarter-factor contraction using four probability arrows and distribution bars under an explicit row-vector convention.
On this page
Build a fictional two-state probability model with four directed conditional transitions, then calculate two-step distributions and stationarity. Probability bars are not observed headcounts or an assumed model of independent repeated choices. Declare states, time steps and matrix orientation before explaining approach through a recurrence.
Declare exhaustive states and a row-vector convention
A and B are mutually exclusive and exhaustive. Given the present state, the next-state distribution does not depend on earlier history, and the same transition rule applies at every step. Use row probabilities p=(a,b), updated by right multiplication p_next=pP. The reference uses column-stochastic matrices/column vectors; transposition reconciles conventions. Do not apply a column formula unchanged to this row table. Create state, arrow, matrix, distribution-bar and condition layers; save native work. p0=(1,0) means certain initial A, not an observed survey headcount.
Check four arrows by their source-state row sums
Set A→A=3/4, A→B=1/4, B→A=1/2, B→B=1/2. These are rows(3/4,1/4),(1/2,1/2) of P, with nonnegative entries summing to 1 per row. Sum outgoing edges for each source state, not incoming edges into A. On a suggested 2200×1600 canvas use centres(650,650),(1550,650), separated reverse arcs and self-loops. Check arrowheads, fraction labels and legend together; colour alone must not identify a state.
Calculate two steps from the certain initial state
p1=(1,0)P=(3/4,1/4). Multiplying again gives A=(3/4)(3/4)+(1/4)(1/2)=11/16 and B=(3/4)(1/4)+(1/4)(1/2)=5/16, still summing to 1. A’s second-step probability is not simply copied as 3/4 because B contributes incoming probability. For total bar width 960px, use 720/240 for p1 and 660/300 for p2, with exact fraction labels. Width maps probability;11/16 does not report eleven observed occurrences out of sixteen.
Independently check two-step B through two paths
Two mutually exclusive paths reach B after two steps: A→A→B with probability(3/4)(1/4)=3/16, and A→B→B with(1/4)(1/2)=1/8, totaling 5/16. Products use conditional transition probabilities, not an assumption that successive states are independent. Draw this path ledger and compare the matrix result; omitting self-loops loses a path. Reusing one matrix over time does not detach transitions from the current state: time homogeneity is not independent identical sampling.
Solve stationarity and substitute both components
Let π=(a,1−a) satisfy πP=π. Then a=(3/4)a+(1/2)(1−a), giving a=2/3 and b=1/3. Substitution yields 1/2+1/6=2/3 for A and 1/6+1/6=1/3 for B, summing to 1. Split the 960px stationary bar 640/320. Stationarity is not the simple row average(5/8,3/8), nor equal transition probabilities. Normalization matters: the delivered vector is a probability distribution, not an arbitrary scalar multiple. At stationarity A→B probability flow is(2/3)(1/4)=1/6 and B→A is(1/3)(1/2)=1/6. An unchanged distribution does not mean every trajectory stays put. Retain self-loops and cross-state edges rather than erase arrows at stationarity.
Prove approach rather than finite-step equality
A probabilities satisfy a_next=1/2+a_current/4, hence a_next−2/3=(a_current−2/3)/4. From a0=1, a_n=2/3+1/(3·4^n). The extra term is positive for every finite nonnegative integer n, so the distribution approaches stationarity without exactly reaching it in finite steps. Distinguish equality from approximation; a displayed 0.667 does not prove exact equality. This recurrence proves convergence for this model, not unconditionally for chains with other rules, absorbing classes or periodicity. Starting at π would keep π at every step, consistent with the strict finite-step deviation from certain A. Deliver the initial condition with the closed form instead of applying that solution unchanged to every starting distribution.
row convention: p_next=p*P
P=((3/4,1/4),(1/2,1/2))
p0=(1,0); p1=(3/4,1/4); p2=(11/16,5/16)
A-A-B=3/16; A-B-B=1/8; total B=5/16
pi=(2/3,1/3); pi*P=pi
a_next=1/2+a/4
a_n=2/3+1/(3*4^n), n=0,1,2,...
finite n: a_n>2/3; limit a_n=2/3
Diagnose invalid rows and transpose mistakes
A mistaken row(3/4,1/2) sums to5/4 and is not a transition distribution; check inputs rather than forcibly renormalize the answer. Reading this matrix unchanged as column-stochastic also fails: its first column sums to5/4. Preserve a one-factor error/control and correction. Diagnose exhaustive states, row sums, arrow directions, vector convention, multiplication and normalization. Bars filling a common width do not prove valid probabilities. This fictional model predicts no real population movement or recommended behaviour.
Reopen and reconstruct matrix and recurrence from arrows
Save editable arcs, labels and three distribution rows; inspect SVG/PDF and a separately opened PNG. At reduced size distinguish self-loops from reverse arrows; enlarge fractions and deviation signs. Reopen, hide the matrix, rebuild rows from outgoing edges, and recalculate p2/stationarity. Deliver row convention, path check, recurrence and invalid-row control, marking unperformed output steps pending. Drawing is manual and matrix/limit work external; infer no App Markov simulator, random experiment or automatic stationary solver.
Related tutorials and sources
VectorPaint two-draw probability trees: connect replacement assumptions to four paths; VectorPaint Bayes source diagram: condition on red within sixty-four area units.
Method sources: Georgia Tech Interactive Linear Algebra, Stochastic Matrices.
Version and operation scope: checked on 18 November 2026 against the official VectorPaint listing and Apple public lookup: version 1.0.8, requiring iOS / iPadOS 17.0 or later. Draw manually with published shapes, pen, Bézier nodes, grids/guides, alignment, layers/groups, text and file output through visible installed controls. Prepare definitions and calculations externally; infer no built-in formula, statistics or graph solver. Keep editable native work and inspect SVG, PDF or PNG as needed. SVG/PDF containing imported bitmaps do not automatically become entirely vector. Canvases and samples are original teaching settings, without physical-device execution or real measurement; mark unperformed actions pending.