Symbolic Operation Ability

In previous chapters, we learned various rules and concepts. In this chapter, we will examine yourproficiency, and see whether you can derive the formulas yourself.

This ability directly determines whether you can follow the derivations in a paper, rather than getting stuck where the author skips steps. Every knowledge point is accompanied by interactive demonstrations, especially the geometric split animation of (a+b)², which makes it clear in one viewing where the "cross term 2ab" comes from.

Why "Symbolic Operation Ability" Matters

When reading papers, you often encounter this situation: the author writes "after simplification, we obtain...", skipping three to five steps in between.

If your symbolic manipulation is not fluent enough, these skipped steps will directly keep you out of the door.

If you are fluent, you can fill in these steps yourself and truly "keep up with" the author's reasoning, rather than passively accepting the conclusion.

For the same rule, "knowing it" and "being able to use it fluently with your eyes closed" are two levels of ability. This chapter trains the latter.

From (a+b)² to Multivariate Expansion

The Most Basic Identity

Geometric Perspective: Why There's an Extra "2ab"

holdThink of it as "the area of a square with side length a+b", and cut this square into 4 pieces:

Adding up the four areas。

This "2ab" doesn't appear out of thin air; it comes from the two crossed rectangles. Click the "Expand" button below to see how the four areas separate.

Generalizing to Three Variables

According to the rule that "every pair of variables produces a cross term", you can directly write:

Derivation logic: treatas, and multiply each term with each term pairwise when expanding.

Multiplying identical variables gives the square term (), multiplying different variables occurs twice (e.g.,and), after combining we get。

Generalizing to Vectors

This rule repeatedly appears in vector operations in another form, e.g., computing the squared norm of the difference of two vectors:

The essence is to replace the"numbers" inside with "vectors", and "multiplication" with "dot product".

When you see a similar expansion in a paper, you can directly recall the rule learned here—new formulas are often old rules in a new guise. This is exactly how "symbolic manipulation ability" helps you "see through" new formulas.

Running the above code outputs:

2 3 True
5 1 True
0.5 2.5 True
-1 4 True
True

This "random substitution verification" method is very practical: when you forget formula details, plugging in a few numbers can confirm whether the version you remember is correct.

Simplification, Transformation, Completing the Square

Completing the Square: Turning the General Form into "Vertex Form"

The goal of completing the square is to turnthis kind of "missing a piece" expression into a complete square:

Example: complete the square forto get。

Verification: expand, they are indeed equal.

Two typical uses of completing the square:

General Approach to Simplification

When encountering a complex expression, usually try in this order:

Example of substitution method: Simplify。

令 t = x² + 1
原式 = t² - 3t + 2 = (t-1)(t-2)
代回 t：
     = (x²+1-1)(x²+1-2)
     = x² · (x²-1)
     = x⁴ - x²

Whether You Can Derive Formulas Yourself Is the Watershed

"Understanding" and "Being Able to Derive" Are Two Different Levels of Ability

Many people get stuck on reading papers, not because of "new knowledge", but because of this level: the formulas use only rules from the first three chapters, but you need to organize the correct sequence of steps yourself without hints.

Complete Derivation Exercise: The Minimum of Mean Squared Error

Problem: Given n data points, find a constant c such that the "sum of squared distances" from all points to cis minimized; find the optimal c.

Derivation process：

设 f(c) = Σᵢ (xᵢ - c)²

展开每一项（用到 (a-b)² 公式）：
f(c) = Σᵢ (xᵢ² - 2cxᵢ + c²)
     = Σᵢ xᵢ² - 2c·Σᵢxᵢ + n·c²    （把 Σ 拆到每一项，c 是常数可以提出来）

这是关于 c 的二次函数，开口向上（c² 系数为 n > 0），
顶点（最小值点）在 c = -B/2A，这里 B = -2Σxᵢ，A = n

代入：c = -(-2Σxᵢ) / (2n) = Σxᵢ / n

Conclusion: The optimal c is exactly themean (average) of all data points.。

This derivation uses only knowledge from this chapter and Chapter 1 (expanding squares, Σ summation, and the vertex formula of quadratic functions), yet it yields the important statistical conclusion of "why the mean is the optimal point that minimizes squared error."

Drag the slider and verify by hand: how the sum of squared distances increases when c deviates from the mean.

Running the above code produces the following output:

5.0
46.0
0 171
3 66
5 46
7 66
10 171

Note that the errors for c=3 and c=7 are equal (both 66) -- they are the same distance from the mean of 5, which reflects the parabola's "symmetry about the vertex."

Chapter Summary