Visual walkthrough — Matrix multiplication — conditions, process, non-commutativity

2.6.5 · D2 · Maths › Matrices & Determinants — Introduction › Matrix multiplication — conditions, process, non-commutativi

Parent note ne tumhe rule bataya tha: do matrices ko multiply karne ke liye, pehli matrix ki rows ko doosri matrix ki columns ke saath dot karo. Yeh page us rule ko ek single idea se derive karta hai — matrices arrows ko idhar-udhar move karti hain — sirf pictures ki madad se. Is page ke end tak tumhe rule yaad karne ki zaroorat nahi hogi; tum use khud rebuild kar sakoge.

Hum assume karte hain ki tum arrows ko tip-to-tail add karna aur stretch karna jaante ho (yeh hai Matrix addition and scalar multiplication). Aur kuch nahi chahiye.

---

Step 1 — Arrow actually hota kya hai? Grid par do numbers

KYA. Flat plane draw karo. Do special chhote arrows chunno: ${\vec{e}}_1$ ek unit right point karta hai, ${\vec{e}}_2$ ek unit upar point karta hai. Hum inhe basis vectors kehte hain — plane ke do "measuring sticks."

KYO. Har doosra arrow inhi do sticks se bani ek recipe hai. Arrow $\vec{x}=\left[\begin{array}{c}2\\ 3\end{array}\right]$ ka literal matlab hai "walk $2$ copies of ${\vec{e}}_1$, phir $3$ copies of ${\vec{e}}_2$." Toh do numbers ka column mysterious nahi hai — yeh ${\vec{e}}_1$ aur ${\vec{e}}_2$ ki language mein likhi hui directions hain.

PICTURE. Neeche, red stick hai ${\vec{e}}_1$, teal stick hai ${\vec{e}}_2$, aur plum arrow hai $\vec{x}=2{\vec{e}}_1+3{\vec{e}}_2$. Dotted path "2 right, 3 up" ki recipe dikhata hai.

---

Step 2 — Ek matrix ek machine hai jo sirf batati hai ki do sticks kahan jaati hain

KYA. Ek transformation koi bhi aisa rule hai jo har arrow ko uthata hai aur kahin naya drop karta hai, lekin grid ko seedha aur evenly spaced rakhta hai (yeh ek linear transformation hai). Yahan ek magic fact hai jis par hum poore page lek lean karenge:

Agar tumhe pata hai ${\vec{e}}_1$ kahan land karta hai aur ${\vec{e}}_2$ kahan land karta hai, toh tumhe pata hai har arrow kahan land karta hai.

KYO. Kyunki $\vec{x}=2{\vec{e}}_1+3{\vec{e}}_2$ hai, aur ek linear machine us recipe ko scramble nahi karta — woh bas har purani stick ko uski nayi landing spot se replace karta hai:

$$T(\vec{x})=2\underset{\text{new right-stick}}{\underbrace{T({\vec{e}}_1)}}+3\underset{\text{new up-stick}}{\underbrace{T({\vec{e}}_2)}}.$$

Har term annotated hai: $2$ aur $3$ unchanged hain (yeh $\vec{x}$ ki apni recipe hai), sirf sticks apne images se swap ho gayi hain.

PICTURE. Left panel: ${\vec{e}}_1,{\vec{e}}_2$ ke saath original grid. Right panel: machine chalane ke baad tilted grid. Sticks ki nayi landing spots matrix ki do columns hain.

---

Step 3 — Matrix ko ek arrow se multiply karna: "columns ko combine karo" rule

KYA. Matrix $B$ (landing spots) aur arrow $\vec{x}=\left[\begin{array}{c}{x}_1\\ x_2\end{array}\right]$ (ek recipe) lo. $B\vec{x}$ find karne ke liye, Step 2 ka fact use karo:

$$B\vec{x}=x_1\underset{\text{where }{\vec{e}}_1\text{ went}}{\underbrace{\left[\begin{array}{c}{B}_{11}\\ B_{21}\end{array}\right]}}+x_2\underset{\text{where }{\vec{e}}_2\text{ went}}{\underbrace{\left[\begin{array}{c}{B}_{12}\\ B_{22}\end{array}\right]}}.$$

YEH HI KYO, AUR KUCH KYO NAHI? Humne $B\vec{x}$ ki definition aise choose ki taaki machine "$B$" recipes ko respect kare. $x_1$ pehle landing column ko scale karta hai, $x_2$ doosre ko scale karta hai, aur hum unhe tip-to-tail add karte hain. Ab us sum ki top row padho:

$$(B\vec{x}{)}_1=B_{11}{x}_1+B_{12}{x}_2.$$

Yeh exactly $B$ ki row 1 ko $\vec{x}$ ke saath dot karna hai. Row·column dot product invent nahi hua — yeh "columns ko combine karo" se apne aap nikal aaya.

PICTURE. Do landing columns scale hote hain (red column times $x_1$, teal column times $x_2$) aur plum result $B\vec{x}$ mein tip-to-tail add hote hain.

---

Step 4 — Do machines ek ke baad ek: isliye $AB$ exist karta hai

KYA. Ab do machines chalao: pehle $B$, phir $A$. $\vec{x}$ feed karo, $B\vec{x}$ milega, phir usse $A$ mein feed karo aur $A(B\vec{x})$ milega. Hum chahte hain ek matrix $C$ jo dono kaam ek saath kare: $C\vec{x}=A(B\vec{x})$. Woh single machine defined hai $C=AB$ ke roop mein.

"$A$ after $B$" ORDER KYO LEKIN LIKHA $AB$? Kyunki hum functions apne input ke left par likhte hain: $A(B\vec{x})$ — jo cheez $\vec{x}$ ke sabse kareeb hai woh pehle act karti hai. Toh $AB$ ka matlab hai "$B$ pehle, phir $A$." Yeh reading non-commutativity ka beej hai: $BA$ par swap karna actions ka order swap kar deta hai, aur joote-phir-moje $\ne$ moje-phir-joote.

PICTURE. Ek teen-panel pipeline: input arrow → $B$ ke baad (tilted grid) → $A$ ke baad (phir tilted). Poore pipeline mein dashed shortcut arrow single machine $C=AB$ hai.

---

Step 5 — $C=AB$ ko column by column banao (poora rule saamne aata hai)

KYA. Step 2 se, hum jaante hain $C$ ko completely agar hum jaante hain $C$ kahan ${\vec{e}}_1$ aur ${\vec{e}}_2$ ko bhejta hai. Toh ${\vec{e}}_1$ ko pipeline se push karo:

$$C{\vec{e}}_1=A\left(B{\vec{e}}_1\right)=A\underset{\text{column 1 of }B}{\underbrace{\left[\begin{array}{c}{B}_{11}\\ B_{21}\end{array}\right]}}.$$

$C$ ka column 1 bas $A$ applied to $B$ ka column 1 hai. Isi tarah ${\vec{e}}_2$ se column 2 milta hai. Toh:

$AB$ ka column $j$ = $A$ times $B$ ka column $j$.

KYO. $B{\vec{e}}_1$ literally $B$ ka column $1$ hai (pehli stick feed karo, us par uski landing spot milti hai). Phir Step 3 batata hai ki $A$ applied to us column ek row·column dot product hai. Step 3 ke dot-product formula se entry $(i,j)$ spell out karte hue:

$$C_{ij}=\underset{\text{row }i\text{ of }A}{\underbrace{A_{i1}}}\underset{\text{col }j\text{ of }B}{\underbrace{B_{1j}}}+A_{i2}{B}_{2j}=\sum _k{A}_{ik}{B}_{kj}.$$

Har symbol earned hai: $A_{ik}$ $A$ ki row $i$ par chalता hai, $B_{kj}$ $B$ ke column $j$ ke neeche jaata hai, aur $k$ woh shared index hai jo unhe milne deta hai.

PICTURE. Final grid $C$ dikhaya gaya hai: column 1 = $A$ acting on $B$ ka column 1, column 2 = $A$ acting on $B$ ka column 2. $(1,1)$ entry highlight hai jahan $A$ ki row-1 $B$ ke column-1 se milti hai.

---

Step 6 — Inner dimensions KYO match KARNI CHAHIYE (degenerate case)

KYA. $B$ ke columns kuch space mein rehne wale arrows hain, maano $n$ numbers ke. Aisa arrow $A$ mein feed karne ke liye, machine $A$ ko $n$-number arrows expect karne chahiye — yaani $A$ ke paas $n$ columns hone chahiye. Agar $A$ ke columns ki sankhya alag hai, toh pipe toot jaata hai.

YEH FAIL KYO HOTA HAI. ${\sum }_k{A}_{ik}{B}_{kj}$ mein, index $k$ ko dono factors mein same range par chalna hai. Agar $A$ ke $2$ columns hain ($k=1,2$) lekin $B$ ke $3$ rows hain ($k=1,2,3$), toh sum mein aisi terms hain jinka koi partner nahi — meaningless. Yeh exactly parent ka Example 3 hai: $A_{2\times 2}$ times $B_{3\times 1}$ undefined hai.

PICTURE. Do connectors: ek matching plug ($A$ ke $n$ columns $B$ ke $n$ rows se fit hote hain, click karte hain) aur ek mismatched plug ($2$-prong socket, $3$-prong plug — physically seat nahi hoga).

---

Step 7 — Order cheezein actually todta hua dekho (non-commutativity, geometrically)

KYA. Maano $A$ ek $90^{\circ }$ rotation hai aur $B$ ek horizontal stretch. Usi chhote L-shape par $B$ phir $A$ ($AB$) aur $A$ phir $B$ ($BA$) karo.

YEH ALAG KYO HOTE HAIN. Ek wide shape ko rotate karne se woh tall ho jaata hai, phir use stretch karne se tall wala choda ho jaata hai. Pehle stretch karne se woh choda ho jaata hai, phir rotate karne se woh differently tall-and-fat ho jaata hai. Alag final shapes ⟹ alag matrices ⟹ $AB\ne BA$. Numbers agree karte hain: $A=\left[\begin{array}{cc}1& 2\\ 3& 4\end{array}\right]$, $B=\left[\begin{array}{cc}0& 1\\ 1& 0\end{array}\right]$ ke saath (parent Example 2), $AB=\left[\begin{array}{cc}2& 1\\ 4& 3\end{array}\right]$ lekin $BA=\left[\begin{array}{cc}3& 4\\ 1& 2\end{array}\right]$.

PICTURE. Do side-by-side pipelines ek hi L se start hote hain, visibly alag L par khatam hote hain.

---

Ek-picture summary

Is page par sab kuch ek sentence hai aur ek picture hai: ek matrix store karti hai ki basis sticks kahan land karti hain; do matrices stack karna matlab hai "land karo, phir dobara land karo," aur use column-by-column padh lena row·column sum ko force karta hai.

Recall Feynman retelling (ek 12-saal ke bachche ko batao)

Do chhote arrows picture karo, ek right point karta hai aur ek upar — yeh tumhare rulers hain. Jo bhi arrow tum draw karo woh bas "itne rights, itne ups" hai. Ek matrix ek cheat-sheet hai: woh sirf tumhe batati hai ki woh do rulers kahan move hote hain; us se tum figure out kar sakte ho ki kuch bhi kahan move karta hai, kyunki har arrow rulers ki motion copy karta hai.

Ab maano tum do moving-machines back to back chalate ho — machine $B$ pehle, machine $A$ doosri. Tum chahte ho ek single combined machine. Use banane ke liye, bas poocho: "right-ruler dono machines ke baad kahan khatam hota hai? up-ruler kahan khatam hota hai?" Woh do answers, side by side, combined machine $AB$ hain. Jab tum uska ek number work out karte ho, tum $A$ ki ek row par aur $B$ ke ek column ke neeche chalne lagte ho, pairs multiply karte ho aur add karte ho — yeh woh famous rule hai, aur ab tum dekh rahe ho ki yeh diya nahi gaya tha, yeh forced tha.

Do loose ends: $B$ se nikalne wale rulers bilkul us size ke hone chahiye jo machine $A$ consume karna jaanti hai, isliye inner dimensions match karni chahiye — warna machines plug together nahi hoti. Aur $B$-phir-$A$ chalana alag journey hai $A$-phir-$B$ se, bilkul jaise moje phir joote pehenna joote phir moje se alag hai — isliye $AB$ aur $BA$ usually disagree karte hain.

Recall

$AB$ ka column $j$ kya equal hota hai? ::: $A$ applied to $B$ ka column $j$. $C_{ij}={\sum }_k{A}_{ik}{B}_{kj}$ forced (chosen nahi) kyun hai? ::: Kyunki hum demand karte hain $C\vec{x}=A(B\vec{x})$; yeh sum wahi hai jo "$B$ ke columns combine karo, phir $A$ chalao" produce karta hai. $A$ ke columns $B$ ke rows ke equal kyun hone chahiye? ::: Shared index $k$ ko dono factors mein same range par chalna hai, warna terms ka koi partner nahi hota. Geometrically, $AB\ne BA$ kyun hai? ::: Woh do motions ko opposite orders mein apply karte hain, aur transformations ka order outcome change karta hai.

---

Connections

• Matrix multiplication — conditions, process, non-commutativity — woh parent rule jise yeh page derive karta hai
• Linear transformations — "moving machine" picture jo poore page mein use ki gayi
• Matrix addition and scalar multiplication — columns ki scaling aur tip-to-tail adding (Steps 1, 3)
• Identity matrix — woh machine jo dono rulers ko wahin chhodti hai
• Inverse of a matrix — woh machine jo doosri ki motion ko undo karti hai
• Matrix transpose properties — $(AB{)}^T=B^T{A}^T$, order bilkul pipeline ki tarah reverse hota hai
• System of linear equations — $A\vec{x}=\vec{b}$ ko "kaun si recipe $\vec{x}$ $\vec{b}$ par land karti hai" ki tarah padhna