Cauchy Principal Value Is Not Ordinary Convergence

Two Different Questions About a Singular Integral

An integrand that becomes unbounded inside an interval cannot be integrated across that point by treating the interval as an ordinary proper integral. Instead, ordinary convergence is defined by approaching the singularity from each side separately. This separate-limit requirement matters for \(1/x\) on \([-1,1]\): its two sides diverge, even though symmetric truncation produces exact cancellation.

The distinction is between two different limiting procedures. The ordinary improper integral asks whether each one-sided contribution has a finite limit on its own. The Cauchy principal value asks what happens when equal lengths are removed on both sides at once. A finite answer to the second question does not answer the first.

Key distinction: A Cauchy principal value is a limit under a specified symmetric truncation. An ordinary improper integral across an interior singularity exists only when the two one-sided improper integrals both converge to finite values.

The Ordinary Integral Across an Interior Singularity

Suppose \(f\) is continuous on each side of an interior point \(c\), but is singular at \(c\), with \(a<c<b\). The ordinary improper integral across \(c\) is defined by splitting at that point and taking two independent one-sided limits:

\int_a^b f(x)\,dx = \lim_{u\to c^-}\int_a^u f(x)\,dx + \lim_{v\to c^+}\int_v^b f(x)\,dx.

This definition gives a finite ordinary improper integral only if both limits exist as finite numbers. One side cannot compensate for divergence on the other: the limits are separate, and the definition does not take a joint limit that links how quickly \(u\) and \(v\) approach \(c\).

For \(1/x\) on \([-1,1]\), the singular point is \(0\). The two required limits are therefore

\lim_{u\to 0^-}\int_{-1}^{u}\frac{1}{x}\,dx \qquad\text{and}\qquad \lim_{v\to 0^+}\int_v^1\frac{1}{x}\,dx.

The antiderivative \(\ln|x|\) can be used on either side of zero, where \(1/x\) is continuous. Evaluate each limit separately; do not first combine the two integrals into one expression with a shared truncation.

Worked Example: The Two One-Sided Integrals of \(1/x\)

On the left, take \(-1<u<0\). The antiderivative gives

\int_{-1}^{u}\frac{1}{x}\,dx = \ln|u|-\ln|-1| = \ln|u|-\ln 1 = \ln|u|.

As \(u\to0^-\), \(|u|\to0^+\), so \(\ln|u|\to-\infty\). Thus the left-hand integral has no finite limit.

On the right, take \(0<v<1\). Then

\int_v^1\frac{1}{x}\,dx = \ln|1|-\ln|v| = \ln 1-\ln v = -\ln v.

As \(v\to0^+\), \(\ln v\to-\infty\), so \(-\ln v\to+\infty\). The right-hand integral also has no finite limit. Since neither one-sided integral converges to a finite number, the ordinary improper integral \(\int_{-1}^1 1/x\,dx\) does not converge. This agrees with the definition: both sides must converge separately.

Symmetric Truncation and the Principal Value

The Cauchy principal value uses a different procedure. For a singularity at \(0\) inside \([-1,1]\), it removes an interval of equal width on both sides, then lets that width shrink to zero. In this case, the definition is

\operatorname{PV}\int_{-1}^{1}\frac{1}{x}\,dx = \lim_{\varepsilon\to0^+} \left( \int_{-1}^{-\varepsilon}\frac{1}{x}\,dx + \int_{\varepsilon}^{1}\frac{1}{x}\,dx \right), \qquad 0<\varepsilon<1.

For each fixed \(\varepsilon>0\), both integrals in parentheses are proper integrals. Their sum is evaluated before taking the limit. This linked, symmetric truncation is what distinguishes the principal value from the ordinary improper integral.

Worked Example: The Cauchy Principal Value of \(1/x\)

Fix \(0<\varepsilon<1\). Evaluate the negative-side integral:

\int_{-1}^{-\varepsilon}\frac{1}{x}\,dx = \ln|-\varepsilon|-\ln|-1| = \ln\varepsilon-\ln 1 = \ln\varepsilon.

Evaluate the positive-side integral:

\int_{\varepsilon}^{1}\frac{1}{x}\,dx = \ln|1|-\ln|\varepsilon| = \ln 1-\ln\varepsilon = -\ln\varepsilon.

The truncated sum is therefore

\int_{-1}^{-\varepsilon}\frac{1}{x}\,dx + \int_{\varepsilon}^{1}\frac{1}{x}\,dx = \ln\varepsilon-\ln\varepsilon = 0.

This sum equals \(0\) for every \(0<\varepsilon<1\), so its limit as \(\varepsilon\to0^+\) is \(0\). Hence the Cauchy principal value exists and equals \(0\). This does not contradict the previous example: the separately defined ordinary integral diverges, while the principal value exists under symmetric truncation.

Why Cancellation Does Not Prove Ordinary Convergence

The reason is built into the definitions. For ordinary convergence, the left and right limits must each settle to finite values. For the principal value, the two contributions are added while their cutoffs remain linked. Large positive and negative quantities can cancel in the linked sum even though neither quantity has a finite limit alone.

In the example, the left contribution is \(\ln\varepsilon\), which tends to \(-\infty\), and the right contribution is \(-\ln\varepsilon\), which tends to \(+\infty\). Their sum is zero for every positive \(\varepsilon\), but this identity does not make either one-sided limit finite.

Derivation of the distinction: The ordinary integral requires the two one-sided limits to exist separately. The principal value instead takes the limit of their sum with equal cutoffs. Since a limit of sums under a linked choice of cutoffs does not imply that either summand has a finite limit, a principal value alone cannot establish ordinary convergence.

One way to see the role of the linked cutoffs is to let the negative and positive truncations have different widths. Write the negative cutoff as \(a>0\) and the positive cutoff as \(b>0\), where \(a<1\) and \(b<1\). The resulting truncated sum is

\int_{-1}^{-a}\frac{1}{x}\,dx + \int_b^1\frac{1}{x}\,dx = \ln a-\ln b = \ln\left(\frac{a}{b}\right).

When \(a=b\), this is zero, giving the symmetric principal value. When \(a\) and \(b\) approach zero at different relative rates, the result need not approach zero or even have a finite limit.

Worked Example: Unequal Cutoffs Give a Different Limit

Choose \(a=2t\) and \(b=t\), where \(0<t<1/2\). Both truncations are valid, and both widths approach zero as \(t\to0^+\). The truncated sum is

\int_{-1}^{-2t}\frac{1}{x}\,dx + \int_t^1\frac{1}{x}\,dx = \ln(2t)-\ln t = \ln\left(\frac{2t}{t}\right) = \ln 2.

The value is exactly \(\ln 2\) for every \(0<t<1/2\), so the limit is \(\ln 2\), not \(0\). If instead \(a=3t\) and \(b=t\), the same calculation gives \(\ln 3\). These are not the principal value, because the cutoffs are not equal. They show why a linked cancellation depends on how the two sides are paired; ordinary convergence cannot be concluded from the symmetric pairing.

A Finite Principal Value Does Not Repair the Separate Limits

The same distinction remains if a regular term is added to the singular integrand. This gives a useful check that the result is about the definitions, not merely about the particular number \(0\). Consider \(1/x+1\). The added constant contributes a finite amount, but it does not change the divergent behavior of the two separate \(1/x\) contributions.

Worked Example: Adding a Constant Changes the Principal Value, Not Convergence

For \(0<\varepsilon<1\), calculate the symmetric truncated sum:

\int_{-1}^{-\varepsilon}\left(\frac{1}{x}+1\right)\,dx + \int_{\varepsilon}^{1}\left(\frac{1}{x}+1\right)\,dx.

An antiderivative on each side is \(\ln|x|+x\). The first integral is

\bigl(\ln\varepsilon-\varepsilon\bigr) - \bigl(\ln 1-1\bigr) = \ln\varepsilon-\varepsilon+1.

The second integral is

\bigl(\ln 1+1\bigr) - \bigl(\ln\varepsilon+\varepsilon\bigr) = 1-\ln\varepsilon-\varepsilon.

Adding these finite truncated values gives

(\ln\varepsilon-\varepsilon+1) + (1-\ln\varepsilon-\varepsilon) = 2-2\varepsilon.

As \(\varepsilon\to0^+\), this tends to \(2\). Thus the principal value is \(2\). But the one-sided integrals still fail to have finite limits: the left side contains \(\ln\varepsilon\to-\infty\), while the right side contains \(-\ln\varepsilon\to+\infty\); the terms involving \(\varepsilon\) and the constant remain finite. Therefore the ordinary improper integral does not converge. The principal value has changed from \(0\) to \(2\), but ordinary convergence has not been established.

A Common Mistake

A tempting but incorrect calculation is to combine the one-sided antiderivative values as if both were ordinary endpoint values at zero:

\int_{-1}^{1}\frac{1}{x}\,dx \overset{\text{incorrect}}{=} \ln|1|-\ln|-1| = 0.

This calculation applies the antiderivative across a point where the integrand is undefined and unbounded. It also skips the separate one-sided limits required by the definition. The corrected calculation evaluates each side independently: the left tends to \(-\infty\), and the right tends to \(+\infty\). Therefore the ordinary integral diverges. The value \(0\) belongs only to the symmetric Cauchy principal value.

Practical check: Before reporting a value for an integral with an interior singularity, ask whether each one-sided improper integral has a finite limit. If only a symmetric truncated sum has a finite limit, identify it as a principal value, not as an ordinary convergent integral.

Why the Distinction Matters

Ordinary improper convergence records whether the integral on each side of the singularity is finite on its own. The principal value records a particular cancellation produced by a particular truncation. Both notions can be useful, but they answer different questions and should be labelled accurately.

For \(1/x\) on \([-1,1]\), the conclusions are therefore unambiguous:

Question	Result
Does the left one-sided improper integral converge to a finite value?	No; it tends to \(-\infty\).
Does the right one-sided improper integral converge to a finite value?	No; it tends to \(+\infty\).
Does the ordinary improper integral across zero converge?	No; both required one-sided limits must be finite.
Does the symmetric Cauchy principal value exist?	Yes; it equals \(0\).

The essential point is not that cancellation is forbidden. Cancellation is exactly what defines the principal value in this example. The point is that cancellation under symmetric truncation does not change the separate-limit definition of ordinary convergence.

Check Your Understanding

Answer each question by distinguishing the separate one-sided limits from the linked symmetric limit.

What two limits must be finite for an ordinary improper integral across an interior singularity to converge?
Evaluate \(\int_{-1}^{-\varepsilon}1/x\,dx+\int_{\varepsilon}^{1}1/x\,dx\). What is its limit as \(\varepsilon\to0^+\)?
Why does the symmetric principal value of \(1/x\) not establish ordinary convergence on \([-1,1]\)?
For unequal cutoffs \(a>0\) and \(b>0\), what is the truncated sum \(\int_{-1}^{-a}1/x\,dx+\int_b^1 1/x\,dx\)?
Is the value obtained by setting both antiderivative endpoints to the singular point a valid evaluation of the ordinary integral? Explain why or why not.