Why Products of Oscillations Need a Bound
A trigonometric factor such as \(\sin(1/x)\) can keep changing as \(x\) approaches zero: its input grows without bound, so the factor need not settle toward a single value. A second factor, such as \(\cos(1/x^2)\), oscillates on a different scale. Trying to track the peaks and troughs of both factors at once is usually unnecessary.
Instead, use a uniform bound. Every sine and cosine value lies between \(-1\) and \(1\), regardless of its input. Their product therefore stays bounded, even when the inputs oscillate at different rates. A power of \(x\) can then shrink the entire product.
Bounding a Product of Trigonometric Factors
For every real number \(u\), \(|\sin u|\leq 1\) and \(|\cos u|\leq 1\). Consequently, for any real-valued inputs \(g_1(x),\ldots,g_m(x)\),
The inputs can have very different forms, such as \(1/x\), \(1/x^2\), or \(1/x^5\). These different frequencies do not change the bounds. More generally, a product that mixes sine and cosine factors also has absolute value at most \(1\).
For \(x>0\), consider
where \(p\) is real and the inputs are defined for \(x\) near \(0\), with \(x\neq 0\). If \(p>0\), then \(x^p\to 0\) as \(x\to 0^+\). The trigonometric factors give
Both bounds tend to zero, so the Squeeze Theorem gives \(f(x)\to 0\). The conclusion depends on the power tending to zero, not on either trigonometric factor having a limit.
Why the Squeeze Argument Works
Here is the full reasoning in terms of the definition of a limit. Let \(\varepsilon>0\). Because \(p>0\), we can choose \(\delta=\varepsilon^{1/p}>0\). If \(0<x<\delta\), then \(x^p<\varepsilon\). Since \(|T(x)|\leq 1\),
Thus, for every \(\varepsilon>0\), choosing this \(\delta\) ensures that \(0<x<\delta\) implies \(|x^pT(x)-0|<\varepsilon\). This is exactly the definition of \(\lim_{x\to 0^+}x^pT(x)=0\). The specific frequencies never enter the argument; the uniform bound on the product is what matters.
Identify the sine and cosine factors, including their inputs.
Use \(|\sin u|\leq 1\) and \(|\cos u|\leq 1\) for every real input \(u\).
For \(x>0\), a positive power gives \(|x^pT(x)|\leq x^p\).
If \(p>0\), the lower and upper bounds, \(-x^p\) and \(x^p\), both tend to zero.
Worked Examples
Worked Example: Two Different Frequencies with a Positive Power
Find \(\displaystyle\lim_{x\to 0^+}x^{1/2}\sin(1/x)\cos(1/x^2)\).
First bound the trigonometric factors separately:
Multiplying these inequalities gives \(|\sin(1/x)\cos(1/x^2)|\leq 1\). Since \(x^{1/2}>0\) for \(x>0\),
As \(x\to 0^+\), both bounds tend to \(0\). Therefore, the Squeeze Theorem gives
Check: The absolute value of the expression is at most \(x^{1/2}\), which tends to zero, so the computed limit is consistent with the bound.
Worked Example: Several Oscillations and a Constant Multiplier
Find \(\displaystyle\lim_{x\to 0^+}4x^{3/4}\sin(3/x)\cos(1/x^2)\sin(1/x^5)\).
Each trigonometric factor has absolute value at most \(1\). Thus, the absolute value of their product is at most \(1\), and
Equivalently, the expression lies between \(-4x^{3/4}\) and \(4x^{3/4}\). Both bounds tend to \(0\), because \(3/4>0\). Hence,
Check: The product of the three trigonometric factors cannot have magnitude greater than \(1\), so the original expression cannot have magnitude greater than \(4x^{3/4}\). That bound shrinks to zero.
What If the Power Is Zero or Negative?
The positive-power result has a necessary role for this particular bound: when \(p=0\), the envelope does not shrink, and when \(p<0\), it grows without bound. A bound alone may not determine a limit in either case. For the specific product \(\sin(1/x)\cos(1/x^2)\), we can show directly that there is no limit by finding sequences approaching zero along which the product approaches different values.
Worked Example: Zero Power Does Not Force a Limit
Consider \(P(x)=\sin(1/x)\cos(1/x^2)\) as \(x\to 0^+\). We construct two sequences with different limiting values of \(P\).
For each positive integer \(n\), let \(A_n=\pi/2+2\pi n\). Choose a positive even integer \(k_n\) that is within \(1\) of \(A_n^2/\pi\); this is possible because consecutive even integers differ by \(2\). Set \(x_n=1/\sqrt{k_n\pi}\). Then \(x_n\to 0^+\) and \(\cos(1/x_n^2)=\cos(k_n\pi)=1\).
Also, \(|k_n\pi-A_n^2|\leq\pi\), so
Since \(\sin(A_n)=1\) and sine is continuous, \(\sin(1/x_n)\to 1\). Therefore \(P(x_n)\to 1\).
Now choose a positive odd integer \(j_n\) within \(1\) of \(A_n^2/\pi\), and set \(y_n=1/\sqrt{j_n\pi}\). The same estimate shows \(1/y_n-A_n\to 0\), so \(\sin(1/y_n)\to 1\). But \(\cos(1/y_n^2)=\cos(j_n\pi)=-1\), giving \(P(y_n)\to -1\).
Both sequences approach \(0\), but the function approaches different values along them. Therefore, \(\lim_{x\to 0^+}P(x)\) does not exist.
Check: The even and odd choices make the cosine factor exactly \(1\) and \(-1\), respectively. In both cases the sine factor tends to \(1\), so the product has the claimed sequence limits.
The same sequences show what happens for negative powers in this example. If \(p<0\), then \(x_n^p\to+\infty\) and \(y_n^p\to+\infty\). Along the first sequence \(x_n^pP(x_n)\to+\infty\); along the second, \(y_n^pP(y_n)\to-\infty\). Thus this product has no finite limit for \(p<0\). This is a conclusion about the displayed example, not a universal conclusion about every expression with a negative power: particular factors might cancel or have additional structure.
A Common Mistake: Assuming Oscillations Cancel
A tempting but invalid step is to assume that rapidly changing factors average out or cancel each other. Oscillating at different rates does not, by itself, imply cancellation, and an expression that is bounded is not necessarily convergent.
Worked Example: Correcting an Invalid Cancellation Argument
Suppose someone claims that \(\sin(1/x)\cos(1/x^2)\to 0\) because both factors oscillate more and more rapidly. The claim does not follow from those facts. The even and odd sequences constructed above make the product tend to \(1\) and \(-1\), respectively. Therefore, it has no limit.
By contrast, for \(x^{1/4}\sin(1/x)\cos(1/x^2)\), the valid calculation is
The factor \(x^{1/4}\), not an assumed cancellation between the oscillations, proves the limit is zero.
Check: Removing \(x^{1/4}\) removes the shrinking bound. The original oscillatory product has subsequential limits \(1\) and \(-1\), while the expression with the positive power is squeezed to \(0\).
Why the Power Matters
The power \(x^p\) acts as an envelope: it limits the largest possible magnitude of the expression. If the trigonometric product is bounded by \(1\), then the entire product has magnitude at most \(x^p\) for \(x>0\). A positive exponent makes that maximum magnitude shrink to zero. Different frequencies can change how often the expression swings between positive and negative values, but they cannot make it exceed the envelope.
This reasoning is especially useful when a direct analysis of the oscillations would be complicated. Do not try to predict where every factor reaches a maximum. First ask whether a simple absolute-value bound already forces the desired limit. The approach also clarifies its limits: if the envelope does not tend to zero, boundedness alone does not establish convergence.
Check Your Understanding
Use absolute-value bounds and sequence reasoning where appropriate.
- Find \(\displaystyle\lim_{x\to 0^+}x^{2/3}\cos(1/x^4)\sin(1/x)\), and state the bound that justifies your answer.
- Find \(\displaystyle\lim_{x\to 0^+}5x^2\sin(7/x)\cos(1/x^3)\), showing the lower and upper bounds.
- Why does \(|\sin(1/x)\cos(1/x^2)|\leq 1\) not prove that this product has a limit?
- What changes in the squeeze argument if the multiplier is \(x^0\) rather than \(x^p\) for \(p>0\)?
- For the specific product \(x^p\sin(1/x)\cos(1/x^2)\), what do the even and odd sequences in the worked example imply when \(p<0\)?
