📚 Diagonalisation of Square Matrices | 方阵的对角化
rtible matrix of eigenvectors and D is a diagonal matrix of eigenvalues, turning difficult matrix multiplication into simple arithmetic on numbers.
对角化是进阶数学中最有力的思想之一。它可以把方阵 A 改写成 A = PDP⁻¹，其中 P 是由特征向量组成的可逆矩阵，D 是由特征值构成的对角矩阵，从而把困难的矩阵乘法变成简单的数字运算。
This topic builds directly on eigenvalues and eigenvectors, and it is especially useful for computing high powers of matrices and for simplifying linear transformations. In AQA Further Mathematics, exam questions usually ask you to construct P and D explicitly, then use them to evaluate a power such as A⁵ or Aⁿ.
本主题直接建立在特征值与特征向量的基础之上，尤其适用于计算矩阵的高次幂以及简化线性变换。在 AQA 进阶数学中，考题通常要求你显式构造 P 和 D，然后利用它们求出 A⁵ 或 Aⁿ 这类幂。
1. What Is Diagonalisation? | 什么是对角化？
P⁻¹. The definition is short, but every part of it matters: P must be invertible, D must be genuinely diagonal (all off-diagonal entries zero), and the product must reproduce A exactly.
若存在可逆矩阵 P 和对角矩阵 D，使得 A = PDP⁻¹，则称方阵 A 可对角化。定义很短，但每一部分都重要：P 必须可逆，D 必须是对角矩阵（非对角元全为零），且乘积必须完全等于 A。
The diagonal entries of D are the eigenvalues of A, and the columns of P are corresponding eigenvectors placed in the same order. If the first diagonal entry of D is λ₁, then the first column of P must be an eigenvector belonging to λ₁, and so on down the diagonal.
D 的对角线元素是 A 的特征值，P 的列是对应的特征向量，且排列顺序与特征值一致。如果 D 的第一个对角元是 λ₁，那么 P 的第一列就必须是属于 λ₁ 的特征向量，依此类推。
nal matrices only requires multiplying the corresponding diagonal entries, so Aⁿ = PDⁿP⁻¹ needs only the nth powers of the individual eigenvalues.
这不仅仅是形式上的改写：它把矩阵乘法，尤其是求幂，转化为简单得多的对角矩阵运算。对角矩阵相乘只需把对应的对角元相乘，因此 Aⁿ = PDⁿP⁻¹ 只需计算各个特征值的 n 次幂。
h shows that A and D are similar matrices. Similar matrices share the same eigenvalues, the same determinant and the same trace, which is exactly why diagonalisation is allowed at all.
这个关系还可以反过来写成 D = P⁻¹AP，说明 A 与 D 是相似矩阵。相似矩阵具有相同的特征值、相同的行列式和相同的迹，这正是对角化能够成立的根据。
2. Prerequisite: Eigenvalues and Eigenvectors | 预备知识：特征值与特征向量
e scalar λ, called the eigenvalue. The vector v must be non-zero, because the zero vector would satisfy the equation trivially for every λ and carry no information.
对于方阵 A，若存在非零向量 v 和标量 λ 使得 Av = λv，则称 v 为特征向量，λ 为特征值。向量 v 必须非零，因为零向量对任何 λ 都平凡地满足该式，不携带任何信息。
rix of the same size. This rearrangement is the key step, because it converts a statement about matrix-vector products into a homogeneous linear system.
该定义式可改写为 (A − λI)v = 0，其中 I 是相同阶数的单位矩阵。这一步改写至关重要，因为它把关于矩阵与向量乘积的陈述转化为齐次线性方程组。
s inverse would force v = 0, contradicting the requirement that eigenvectors are non-zero.
非零解 v 存在当且仅当矩阵 A − λI 是奇异矩阵，因此其行列式必须为零。如果 A − λI 可逆，两边左乘其逆矩阵就会推出 v = 0，与特征向量非零的要求矛盾。
This link between eigenvalues and determinants is the foundation of the characteristic equation. It also explains why diagonalisation depends so heavily on being able to find enough independent eigenvectors: each eigenvector supplies one column of P.
特征值与行列式之间的这种联系是特征方程的基础。它也解释了为什么对角化如此依赖于找到足够多线性无关的特征向量：每一个特征向量都会提供 P 的一列。
Note that any non-zero multiple of an eigenvector is still an eigenvector for the same eigenvalue. This freedom is useful in exams because you can scale eigenvectors to avoid fractions when building P.
注意，特征向量的任意非零倍数仍是同一特征值对应的特征向量。这种自由度在考试中很有用，因为你可以缩放特征向量，从而在构造 P 时避免分数。
3. The Characteristic Equation | 特征方程
panding the determinant produces a polynomial in λ, and its roots are precisely the eigenvalues of A.
要求特征值，需解特征方程 det(A − λI) = 0。展开行列式会得到关于 λ 的多项式，其根恰好就是 A 的特征值。
an n × n matrix has exactly n eigenvalues when counted with multiplicity over the complex numbers.
对于 2×2 矩阵，这会得到关于 λ 的二次方程；对于 3×3 矩阵，则会得到关于 λ 的三次方程。多项式的次数总等于矩阵的阶数，因此 n×n 矩阵在复数范围内按重数计算恰好有 n 个特征值。
The roots are the eigenvalues, and each root' as algebraic multiplicity 2, and that repetition has real consequences for diagonalisability.
这些根就是特征值，每个根的代数重数是指它作为因式出现的次数。例如重根 (λ − 3)² 意味着 λ = 3 的代数重数为 2，这种重复性对能否对角化有实质影响。
| Matrix size | Characteristic polynomial | Shortcut check |
| 2 × 2 | λ² − (trace)λ + det(A) = 0 | Sum of eigenvalues = trace, product = determinant |
| 3 × 3 | Cubic in λ, usually one rational root found by inspection | Still sum = trace, product = determinant |
because it lets you check your eigenvalues in seconds. If your roots do not add up to the trace, you have made an arithmetic slip somewhere.
对于 2×2 矩阵，值得记住快捷式 λ² − (trace)λ + det(A) = 0，因为它能让你在几秒内检验特征值。如果求得的根之和不等于迹，那么你在某处出现了计算错误。
4. When Is a Matrix Diagonalisable? | 矩阵何时可对角化？
d only if it has n linearly independent eigenvectors. This is the single most important criterion, and it can be tested before you attempt to build P.
n×n 矩阵可对角化，当且仅当它有 n 个线性无关的特征向量。这是最重要的判据，而且可以在动手构造 P 之前先行检验。
If all n eigenvalues are distinct, the matrix is automatically diagonalisable. Eigenvectors belonging to different eigenvalues are always linearly independent, so n different eigenvalues guarantee n independent directions.
如果 n 个特征值互不相同，矩阵必然可对角化。属于不同特征值的特征向量总是线性无关的，因此 n 个不同的特征值保证有 n 个独立方向。
If an eigenvalue is repeated, you must check its geometric multiplicity, meaning the number of independent eigenvectors you can actually find for it. The matrix is diagonalisable only when the geometric multiplicity equals the algebraic multiplicity for every repeated eigenvalue.
如果某个特征值重复出现，就必须检查它的几何重数，也就是你实际能为它找到的线性无关特征向量的个数。只有当每个重特征值的几何重数都等于其代数重数时，矩阵才可对角化。
= 0 gives just one independent eigenvector, so it is defective and cannot be diagonalised.
一个典型的反例是元素为 1, 1, 0, 1 的剪切矩阵。它唯一的特征值是 1，代数重数为 2，但解 (A − I)v = 0 只能得到一个线性无关的特征向量，因此它是亏损的，不能对角化。
Real matrices may also have complex eigenvalues, such as a rotation matrix, and then no real eigenvector basis exists. Over the complex numbers such a matrix is still diagonalisable, but in an AQA Further Mathematics exam you are normally working over the reals.
实矩阵也可能有复特征值，例如旋转矩阵，此时不存在实的特征向量基。在复数范围内这样的矩阵仍可对角化，但在 AQA 进阶数学考试中，你通常是在实数范围内讨论。
5. Constructing P and D | 构造 P 和 D
hen place the eigenvectors as columns of P in the same order as the eigenvalues down the diagonal of D.
构造过程遵循固定套路。先由 det(A − λI) = 0 求出特征值，再对每个特征值解 (A − λI)v = 0 求出特征向量，然后把特征向量作为 P 的列，其顺序与 D 对角线上特征值的顺序一致。
The order of the columns of P is your choice, but it must match the order of the diagonal entries of D exactly. Swapping two columns of P without swapping the corresponding eigenvalues in D gives a completely wrong answer.
P 的列顺序可以自行选择，但必须与 D 的对角元顺序完全对应。如果交换了 P 的两列却没有同时交换 D 中相应的特征值，结果就完全错误。
Each eigenvector may be scaled by any non-zero constant. Choosing integer eigenvectors with small entries, or dividing through by a common factor, keeps the later matrix multiplication manageable and reduces the risk of arithmetic errors.
每个特征向量都可以乘以任意非零常数。选择元素较小、取整数的特征向量，或者约去公因子，可以让后续的矩阵乘法更容易处理，并降低计算出错的风险。
ies, but P then has three columns and D has three diagonal entries. Always verify that det(P) ≠ 0 before going further, because a zero determinant means your eigenvectors were not independent.
对于 3×3 矩阵，套路相同，但此时 P 有三列，D 有三个对角元。继续计算之前务必要检验 det(P) ≠ 0，因为行列式为零意味着你选取的特征向量并不线性无关。
6. Verifying the Factorisation and Finding P⁻¹ | 验证分解与求逆矩阵 P⁻¹
s step of inverting P, and it catches order mistakes immediately, because AP uses A times each eigenvector while PD uses each eigenvalue times each eigenvector.
一个快速的验证方法是检查 AP = PD。这样可以避免更繁琐的求逆步骤，并且能立刻发现顺序错误，因为 AP 是 A 乘以每个特征向量，而 PD 是每个特征值乘以对应的特征向量。
d, −b; −c, a], provided ad − bc ≠ 0. This closed form is quick and reliable, and examiners expect you to use it rather than row reduction.
对于元素为 a, b, c, d 的 2×2 矩阵 P，当 ad − bc ≠ 0 时，其逆矩阵为 P⁻¹ = (1 ÷ (ad − bc)) × [d, −b; −c, a]。这个公式快捷可靠，考官期望你直接使用它，而不是用行变换。
the adjoint method or row reduction to find P⁻¹. Check your work by multiplying P⁻¹P and confirming that you obtain the identity matrix with ones on the leading diagonal and zeros elsewhere.
对于 3×3 矩阵，通常需要用伴随矩阵法或行变换求 P⁻¹。检验的方法是计算 P⁻¹P，确认得到主对角线为 1、其余元素为 0 的单位矩阵。
This final check takes a few lines and is well worth the time, because a single sign error in P⁻¹ will corrupt every later power you compute.
一旦逆矩阵正确，就可以直接检验等式 A = PDP⁻¹。这个最终检验只需几行，非常值得花时间，因为 P⁻¹ 中一个符号错误就会毁掉之后计算的每一个幂。
7. Computing High Powers of a Matrix | 计算矩阵的高次幂
P⁻¹, because the inner P⁻¹P products collapse to the identity matrix, leaving only the two outer copies of P.
对角化最主要的实用价值在于幂公式。把 A = PDP⁻¹ 取 n 次幂得到 Aⁿ = PDⁿP⁻¹，因为中间的 P⁻¹P 乘积都化为单位矩阵，只剩下最外侧的两个 P。
Since D is diagonal, Dⁿ is found by raising each diagonal entry to the power n separately. There is no matrix multiplication involved at this stage, which is why diagonalisation turns an O(n) chain of matrix products into a single cheap calculation.
由于 D 是对角矩阵，Dⁿ 只需把每个对角元分别取 n 次幂即可。这一步不涉及任何矩阵乘法，这正是对角化能把一连串矩阵乘积转化为一次简单计算的原因。
Aⁿ = P diag(λ₁ⁿ, λ₂ⁿ, …, λₖⁿ) P⁻¹
genvalues, we have A⁻¹ = PD⁻¹P⁻¹, where D⁻¹ has diagonal entries 1 ÷ λ₁, 1 ÷ λ₂ and so on.
类似的技巧也能给出 A 的逆。由于 D 是对角矩阵且其对角元是非零特征值，我们有 A⁻¹ = PD⁻¹P⁻¹，其中 D⁻¹ 的对角元为 1 ÷ λ₁、1 ÷ λ₂ 等等。
valuate A⁵. You are expected to compute D⁵ first, then perform the two matrix multiplications carefully, keeping exact fractions rather than decimals.
考试中，题目措辞通常很明确：证明 A = PDP⁻¹，并由此计算 A⁵。你应当先算出 D⁵，再仔细完成两次矩阵乘法，并使用精确分数而不是小数。
8. Applications to Recurrence Relations | 在递推关系中的应用
ⁿ u₀, and once A is diagonalised the long-term behaviour is read off from the eigenvalues.
对角化也是求解耦合递推关系的标准工具。形如 u₍ₙ₊₁₎ = A uₙ 的方程组可以紧凑地写成 uₙ = Aⁿ u₀，而一旦把 A 对角化，长期行为就可以直接从特征值读出。
The eigenvector basis decouples the system. Each coordinate in the eigenbasis simply multiplies by its own eigenvalue at every step, so the growth or decay of the whole system is governed by the size of the largest eigenvalue in absolute value.
特征向量基使方程组解耦。在特征基下，每个坐标每步只需乘以它对应的特征值，因此整个系统的增长或衰减由绝对值最大的特征值决定。
If every eigenvalue satisfies |λ| < 1, the sequence tends to the zero vector. If some eigenvalue satisfies |λ| > 1, the sequence typically grows without bound unless the initial vector has no component in that eigendirection.
如果所有特征值都满足 |λ| < 1，序列趋于零向量。如果某个特征值满足 |λ| > 1，除非初始向量在该特征方向上没有分量，否则序列通常会无界增长。
This is exactly the mathematics behind population models, Markov chains and difference equations, where the dominant eigenvalue determines the steady state and the other eigenvalues determine the transient behaviour.
这正是种群模型、马尔可夫链和差分方程背后的数学原理：主特征值决定稳态，其余特征值决定暂态行为。
9. Diagonalisation and Linear Transformations | 对角化与线性变换
A matrix represents a linear transformation, and in the eigenvector basis that transformation becomes extremely simple. Diagonalisation says the same transformation is just a stretch along each eigenvector direction, with factor equal to the corresponding eigenvalue.
矩阵表示一种线性变换，而在特征向量基下，这种变换变得极其简单。对角化说明同一个变换只是在每个特征向量方向上做伸缩，伸缩倍数等于对应的特征值。
A negative eigenvalue indicates a reflection combined with a stretch along that direction, and an eigenvalue of zero indicates that the transformation collapses that direction entirely, which is why a zero eigenvalue makes the matrix singular.
负特征值表示在该方向上除了伸缩还包含反射，而特征值为零表示变换把该方向完全压缩掉，这正是零特征值会使矩阵奇异的原因。
The change of basis is performed by P, which converts coordinates in the eigenvector basis into the standard basis. So PDP⁻¹ means: change to the eigenbasis, apply the simple diagonal scaling, then change back.
基变换由 P 完成，它把特征向量基下的坐标转换为标准基下的坐标。因此 PDP⁻¹ 的含义是：换到特征基，做简单的对角伸缩，再换回来。
This geometric reading is valuable in longer questions on transformations, where you may be asked to describe the effect of A on the plane or to interpret a determinant as an area scale factor.
这种几何理解在较长的变换题中很有价值，题目可能要求你描述 A 对平面的作用，或把行列式解释为面积伸缩因子。
10. Ordering, Scaling and Similarity | 顺序、缩放与相似性
mns with the matching permutation of the eigenvalues changes D, yet A = PDP⁻¹ still holds.
分解并不唯一。对某个特征向量列进行任意非零缩放会
更多咨询请联系16621398022（同微信）
或截图后用「扫一扫 → 相册」选择本图