Triangular Coordinate Patterns and Independence

Reading Independence from a Triangular Pattern

When a list of vectors is placed into a matrix as columns, its independence is determined by the solutions of a homogeneous system. Sometimes the coordinate matrix has a pattern that makes a full row reduction unnecessary. In an upper-triangular matrix, each row equation involves a coefficient and only the coefficients to its right. Starting with the last equation and working upward can therefore settle independence directly.

Throughout, the vectors and scalars are in \(\mathbb{R}^n\) and \(\mathbb{R}\), respectively. For a list \(\mathbf{v}_1,\dots,\mathbf{v}_n\), put the vectors into the columns of the square matrix \(\mathbf{A}=\begin{bmatrix}\mathbf{v}_1&\cdots&\mathbf{v}_n\end{bmatrix}\). The list is linearly independent exactly when the equation \(\mathbf{A}\mathbf{c}=\mathbf{0}\), where \(\mathbf{c}=(c_1,\dots,c_n)^{\top}\), has only the solution \(\mathbf{c}=\mathbf{0}\). This follows directly from the definition of linear independence.

For an upper-triangular matrix, row \(i\) of \(\mathbf{A}\mathbf{c}=\mathbf{0}\) has the form \(a_{ii}c_i+a_{i,i+1}c_{i+1}+\cdots+a_{in}c_n=0\). Thus, if every diagonal entry is nonzero, the last equation forces \(c_n=0\). The equation above it then forces \(c_{n-1}=0\), and this continues upward. The following result makes the test precise.

Proof. Suppose \(\mathbf{A}\mathbf{c}=\mathbf{0}\). The last row equation is \(a_{nn}c_n=0\). Since \(a_{nn}\ne0\), it follows that \(c_n=0\). Now suppose, for some \(i<n\), that \(c_{i+1}=\cdots=c_n=0\). Row \(i\) gives \(a_{ii}c_i=0\), because every term involving a later coefficient vanishes. Since \(a_{ii}\ne0\), we get \(c_i=0\). Applying this argument successively for \(i=n-1,n-2,\dots,1\) gives \(\mathbf{c}=\mathbf{0}\). The homogeneous system therefore has only the trivial solution, so the columns are independent. \(\square\)

Solving an Upper-Triangular System from the Bottom

The procedure in the proof is backward substitution: use the last row first, then move to the row immediately above it. At each stage, the coefficients already found to be zero remove the later terms from the current equation. The order matters; beginning with the first row may leave several unknown coefficients mixed together. A worked calculation shows how the pattern matches the definition.

The test also handles parameters: an upper-triangular pattern remains useful even when one diagonal entry depends on a scalar. When that entry vanishes, the equations may permit a nonzero coefficient vector. The next result shows that for a square triangular matrix, this is not just a limitation of the test: a zero on the diagonal really does mean dependence.

Proof. The proposition proves that nonzero diagonal entries imply independence. For the converse, we prove by induction on \(n\) that if an upper-triangular matrix has a zero diagonal entry, then its columns are dependent. When \(n=1\), a zero diagonal entry makes the matrix \([0]\), whose single column is the zero vector, so its columns are dependent.

Now let \(n>1\), and write the matrix in block form as

where \(\mathbf{r}\) is a row and \(\mathbf{B}\) is an \((n-1)\times(n-1)\) upper-triangular matrix. If \(a_{11}=0\), the first column of \(\mathbf{A}\) is \(\mathbf{0}\), so the columns are dependent. If \(a_{11}\ne0\), the assumed zero diagonal entry must occur in \(\mathbf{B}\). By the induction hypothesis, there is a nonzero vector \(\mathbf{y}\) with \(\mathbf{B}\mathbf{y}=\mathbf{0}\). Define \(x_1=-\mathbf{r}\mathbf{y}/a_{11}\) and \(\mathbf{x}=(x_1,\mathbf{y}^{\top})^{\top}\). Then \(\mathbf{x}\ne\mathbf{0}\), since \(\mathbf{y}\ne\mathbf{0}\), and block multiplication gives

Thus the homogeneous system has a nonzero solution, so the columns are dependent. This proves that independence requires every diagonal entry to be nonzero, completing both directions. \(\square\)

Examples with Parameters and a Different Orientation

A diagonal entry that varies with a parameter identifies exactly where the triangular independence test changes its conclusion. At the exceptional value, it is useful to produce and check a specific dependence relation rather than stop at the test.

For a lower-triangular matrix, the equations have the reverse pattern: row \(i\) involves \(c_1,\dots,c_i\). With nonzero diagonal entries, one starts at the first row and works downward. This is the same coordinate reasoning in the opposite direction.

Common Mistake: Starting at the Wrong End

For an upper-triangular matrix, using the first row first does not by itself determine the first coefficient, because that row can involve all the coefficients to its right. For example, consider the equation \(\mathbf{A}\mathbf{c}=\mathbf{0}\) with

A wrong calculation starts with the first row, chooses \(c_1=1\), and uses \(c_1+2c_2=0\) to obtain \(c_2=-\tfrac12\). But this proposed vector does not solve the system: the second row gives \(3c_2=-\tfrac32\ne0\). The correction is to start with the last row. It forces \(3c_2=0\), so \(c_2=0\); the first row then forces \(c_1=0\). The order of substitution is part of the method, not a cosmetic choice.

The diagonal test is especially useful when the zero pattern is already visible in the coordinate matrix: it can decide independence without calculating a reduced row echelon form. It is specific to triangular square matrices. A nonzero diagonal in a general, non-triangular matrix is not by itself enough to apply this argument, because the equations may mix coefficients without the one-directional pattern used in substitution. For a triangular matrix with a zero diagonal entry, the theorem gives the opposite conclusion: its columns are dependent.