When the Usual Slope Formula Breaks Down
An implicit equation \(F(x,y)=0\) can describe a curve without expressing \(y\) explicitly as a function of \(x\). At a point where the curve is locally a differentiable graph \(y=y(x)\), implicit differentiation gives the familiar slope formula \(dy/dx=-F_x/F_y\), provided \(F_y\ne 0\). But the denominator can vanish, or both partial derivatives can vanish. These cases are not interchangeable: one can indicate a vertical tangent, while the other leaves the local geometry undetermined.
The key is to distinguish a failure of this particular formula from a failure of the curve itself to be smooth. Partial derivatives describe first-order change in the defining function \(F\). If both vanish, that first-order information disappears; the equation’s nearby branches must then be examined directly.
What the Partial Derivatives Tell Us
Assume \(F\) has continuous first partial derivatives near \((a,b)\) and \(F(a,b)=0\). If \(F_y(a,b)\ne0\), the implicit function theorem guarantees that, near this point, the curve is the graph of a unique continuously differentiable function \(y=y(x)\). Differentiating \(F(x,y(x))=0\) by the chain rule gives
Therefore, wherever \(F_y\ne0\) along this graph,
If instead \(F_y(a,b)=0\) but \(F_x(a,b)\ne0\), the point is still regular. The implicit function theorem now gives a local graph \(x=x(y)\), with
At the point, this derivative is zero. The curve has a vertical tangent there: as \(y\) changes, its first-order change in \(x\) is zero. The formula for \(dy/dx\) cannot be used because its denominator is zero, but this is not a singular point.
Why the Formula Gives No Answer at a Singular Point
At a regular point, the gradient \((F_x,F_y)\) is nonzero and is perpendicular to the curve’s tangent direction. Thus a tangent vector \((u,v)\) satisfies \(F_xu+F_yv=0\). For a graph with tangent vector \((1,dy/dx)\), this condition gives the slope formula when \(F_y\ne0\). At a singular point, however, both partial derivatives vanish, so the same condition becomes \(0=0\) for every \((u,v)\). It cannot select a tangent direction.
This explains the failure precisely: the equation obtained from first-order change is no longer informative. It does not prove that the curve has no tangent, nor that it has one unique tangent. Higher-order terms or a direct analysis of the nearby points may reveal a cusp, multiple branches, an isolated point, or even a smooth curve whose defining equation has a singular gradient.
A Vanishing Denominator at a Regular Point
Worked Example: A Vertical Tangent Without a Singular Point
Consider \(F(x,y)=x+y^2\) at \((0,0)\). First, the point lies on the curve because
The partial derivatives are \(F_x=1\) and \(F_y=2y\), so \(F_x(0,0)=1\) and \(F_y(0,0)=0\). The point is regular, not singular. Solving for \(x\) gives the local curve exactly:
Thus its tangent is vertical, with equation \(x=0\). The slope formula for \(dy/dx\) would require \(-1/0\), which is undefined. The curve also cannot be represented as a single-valued \(y(x)\) on a full interval around \(x=0\): for \(x<0\), it has two values \(y=\pm\sqrt{-x}\). Checking by substitution, \(x=-y^2\) gives \(F(x,y)=-y^2+y^2=0\), and differentiating \(x=-y^2\) confirms the vertical tangent at the origin.
Singular Points Can Have Different Geometries
When both partial derivatives vanish, there is no general rule that identifies the curve’s local shape. Factorization and parameterizations can expose the nearby branches. Two important possibilities are a cusp, where branches meet in a pointed shape, and a crossing, where distinct smooth branches pass through the same point.
Worked Example: A Cusp with a Horizontal Tangent
Consider \(F(x,y)=y^2-x^3\) at the origin. The point lies on the curve, and
So the origin is singular. For \(x\ge0\), the equation gives two branches \(y=x^{3/2}\) and \(y=-x^{3/2}\), meeting at the origin. For \(x<0\), there are no real points on the curve, since \(x^3<0\) while \(y^2\ge0\). On the two branches for \(x>0\), differentiation gives
Both slopes approach \(0\) as \(x\) approaches \(0\) from the right, so the two branches have the same limiting tangent, \(y=0\). Direct substitution checks either branch: \(y^2=(x^{3/2})^2=x^3\), and likewise for its negative. Implicit differentiation gives \(2y\,dy/dx=3x^2\), which at the origin reduces to \(0=0\); it cannot recover the branch slopes there. The cusp has a horizontal limiting tangent, but the relation as a whole is not a single-valued graph \(y(x)\) on a neighborhood of the origin.
Worked Example: Intersecting Branches with Different Slopes
Now let \(F(x,y)=y^2-x^2\). At \((0,0)\), \(F=0\), \(F_x=-2x\), and \(F_y=2y\), so both partial derivatives vanish. Factoring the equation reveals its local branches:
These are two lines crossing at the origin. Their slopes are respectively \(1\) and \(-1\), as differentiation confirms: \(d(x)/dx=1\) and \(d(-x)/dx=-1\). Substitution verifies both branches, since \(x^2-x^2=0\) and \((-x)^2-x^2=0\). At the origin, implicit differentiation yields \(2y\,dy/dx-2x=0\), again just \(0=0\). There is no single tangent slope for the whole crossing: each branch has its own tangent line.
When There Is No Nearby Curve
Worked Example: A Singular Point That Is Isolated
Consider \(F(x,y)=x^2+y^2\). Its only real zero is the origin, because both squares are nonnegative and their sum can be zero only if \(x=y=0\). At that point, \(F_x=2x=0\) and \(F_y=2y=0\), so it is singular. In every punctured neighborhood of the origin, \(x^2+y^2>0\); there are no other points of the level curve approaching it. Therefore there is no curve branch and no tangent line to determine. Substitution confirms \(F(0,0)=0\), while any other point has \(F(x,y)>0\).
The examples show why a singular point cannot be classified from the two first partial derivatives alone. A cusp has branches with a shared limiting tangent; a crossing has distinct tangent directions; and \(x^2+y^2=0\) has no nearby curve at all.
The Defining Equation Matters
Singularity as defined here concerns the function \(F\), not just the geometric set of points where \(F=0\). An equation can have zero gradient along a curve that is itself smooth. For example, \(F(x,y)=(y-x^2)^2\) has the same zero set as \(y=x^2\), a smooth parabola. Yet both partial derivatives of \(F\) vanish at every point of that zero set.
Indeed, \(F_x=-4x(y-x^2)\) and \(F_y=2(y-x^2)\); both are zero whenever \(y=x^2\). Nevertheless the curve can be written explicitly as \(y=x^2\), whose derivative is \(dy/dx=2x\). This example warns against concluding that the geometric curve itself must be cusped or crossed whenever the gradient of one defining function vanishes.
A Practical Way to Analyze Failure
Verify that \(F(a,b)=0\); otherwise it is not on the level curve under study.
If \(F_y\ne0\), use the local graph \(y(x)\) and the slope formula. If \(F_y=0\) but \(F_x\ne0\), solve locally for \(x(y)\); the tangent is vertical at the point.
Try factoring it, solving for one variable, or finding a parameterization to identify real branches near the point.
Determine whether they approach with one tangent, different tangents, or no nearby points. Verify each proposed branch in the original equation.
This procedure matters because the implicit slope formula is a local result with specific hypotheses, not a device that assigns a slope at every point of every equation. A zero denominator calls for the other variable as a local coordinate when the point is regular. A zero-over-zero expression at a singular point calls for examining the actual local branches instead.
Check Your Understanding
For each question, distinguish what the partial derivatives establish from what still requires analysis of the curve.
- What condition makes a point on \(F(x,y)=0\) singular?
- If \(F_y(a,b)=0\) but \(F_x(a,b)\ne0\), what local representation is guaranteed and what is the tangent direction?
- Why does implicit differentiation provide no slope at a point where both partial derivatives vanish?
- How do the tangent behaviors at the cusp \(y^2=x^3\) and the crossing \(y^2=x^2\) differ?
- Why does a vanishing gradient not by itself prove that the zero set has a cusp or crossing?
