Saddle-Node, Transcritical, and Pitchfork Bifurcations

The panorama article promised that when a real eigenvalue crosses zero, only three things can generically happen. This article delivers the three, in full working detail, on the simplest possible stage: a flow on the line, $\dot{x}=f(x,\mu)$ . Everything is visible here. Fixed points are the crossings of the graph of $f$ with the axis; stability is the slope at the crossing; and a bifurcation is what happens to the crossings as the knob $\mu$ slides the curve around. Lift a downward parabola until its peak touches the axis and two crossings merge and vanish — the saddle-node. Let two crossings slide through each other and swap slopes — the transcritical. Flatten a symmetric S-curve until one crossing becomes three — the pitchfork. Three cartoons, but they carry a theory. Near any degenerate point a smooth change of variables crushes every one-parameter family onto one of the canonical polynomials $\dot{x}=\mu-x^2$ , $\dot{x}=\mu x-x^2$ , $\dot{x}=\mu x\mp x^3$ — which is why a neuron's firing threshold, a laser's turn-on, and a column's buckling obey, near threshold, literally the same equation. The fine print is where the physics lives: the pitchfork exists only under an exact symmetry and unfolds spectacularly under the smallest imperfection; the subcritical variant hides jumps and hysteresis inside an innocent sign change; and a vanished saddle-node leaves behind a ghost — a bottleneck that delays the escaping system by a divergent time $\pi/\sqrt{\mu}$ . Master these three and the rest of bifurcation theory is bookkeeping: the center manifold reduces every higher-dimensional static bifurcation to this article.

Key takeaways

On the line, a bifurcation requires $f(x^*,\mu_c)=0$ and $f'(x^*,\mu_c)=0$ : hyperbolic fixed points are protected by the implicit function theorem, so only degenerate crossings can reorganize.
Saddle-node $\dot{x}=\mu-x^2$ : two fixed points ( $\pm\sqrt{\mu}$ , upper stable) exist for $\mu>0$ and annihilate at the fold — the generic case, requiring no structure at all.
Transcritical $\dot{x}=\mu x-x^2$ : the branches $x^*=0$ and $x^*=\mu$ persist for all $\mu$ and exchange stability at the crossing — the shape of every threshold with a state that always exists (laser off, disease-free).
Pitchfork $\dot{x}=\mu x\mp x^3$ : under inversion symmetry a state splits into a mirror pair; the cubic sign separates the soft supercritical branch $\pm\sqrt{\mu}$ from the hard subcritical case with jumps and a hysteresis window.
An imperfection $h$ unfolds the pitchfork into a smooth branch plus a disconnected fold; the bistable region is the cusp $h^2=4\mu^3/27$ , so the perfect fork is a symmetry-protected accident.
Past a vanished fold the flow bottlenecks through the ghost: passage time $T=\pi/\sqrt{\mu}$ , the square-root delay behind type-I intermittency and sluggish tipping points.
Key idea: a local bifurcation happens exactly where a fixed point loses hyperbolicity ( $f=0$ and $f'=0$ together); which normal form applies is dictated by the symmetry of $f$ , and the sign of the leading surviving nonlinearity decides soft versus hard.

The core insight

A local bifurcation happens exactly where a fixed point loses hyperbolicity ( $f=0$ and $f'=0$ together); which of the three normal forms applies is dictated by the symmetry of $f$ and the sign of the leading surviving nonlinearity.

Where this lives today

The three normal forms are standard instruments across quantitative science. Epidemic thresholds are transcritical — the disease-free state exists at every contact rate, and $R_0=1$ is where it swaps stability with the endemic branch — so threshold public health inherits equation (5), imperfections included (case importation plays the role of $h$ , rounding the corner as spontaneous emission rounds a laser threshold). Power-system engineers certify grids against voltage collapse by computing the saddle-node of the load-flow equations and its distance from the operating point; the ghost delay of equation (4) is why post-fault recovery can hang for seconds before collapsing. The subtler predictions earn their keep too. The $\mu^{-1/2}$ bottleneck law identifies saddle-node mechanisms experimentally — in the type-I intermittency of lasers and convection cells, in neuronal spike-timing near threshold, and in ecology's "extinction debt," where populations linger near a vanished equilibrium long after conditions turned fatal. Cusp geometry organizes bistable perception experiments and MEMS switch calibration. And the imperfect pitchfork is a precision tool: cold-atom groups have reconstructed the full cusp $h^2=4\mu^3/27$ from spinor condensates crossing a symmetry-breaking transition, a deliberate microgauss field bias selecting the branch — Landau's 1937 diagram, drawn one atom cloud at a time.

The physical picture

Draw $f(x)$ against $x$ and put arrows on the $x$ -axis: rightward where the curve is above the axis, leftward below. Fixed points are the crossings; a downward slope means the arrows converge (stable), an upward slope means they flee (unstable). Now turn the knob. Every static bifurcation in nature is one of three moves this picture can make.

Figure

The saddle-node mechanism — lifting a downward parabola through the axis makes two fixed points merge into a half-stable point and vanish

The names come from two dimensions, where the colliding pair is a saddle and a node, but the mechanism is purely one-dimensional: a fold of the curve of equilibria. Note what the middle panel shows — at exactly $\mu=0$ the merged point is half-stable, attracting from the right, repelling to the left. That tangency is the geometric meaning of the degeneracy condition $f'=0$ . The transcritical move needs a reason for a crossing to stay pinned to the axis while the curve tilts. Physics supplies such reasons wholesale: a laser with zero photons stays at zero photons; a population of zero individuals stays extinct; a non-rotating state stays non-rotating. Whenever a state exists identically in the parameter, branches can only cross and trade stability — nothing can be created or destroyed. The pitchfork needs more: a full mirror symmetry $x\to-x$ of the dynamics, which forbids every even term in $f$ . Then the symmetric state cannot tilt — it can only shed a pair of mirror-image children. Whether the children are born stable (supercritical, soft) or the parent dies violently (subcritical, hard) is decided by the sign of the cubic — invisible to any linear analysis.

Figure

The three bifurcation diagrams — fold, exchange of stability, and symmetric splitting with solid stable and dashed unstable branches

Two more pictures complete the story, and both carry serious physics. First, break the pitchfork's symmetry by a whisper $h$ : the fork shatters into one smooth branch and a separate fold — the imperfect bifurcation, drawn below, with a cusp-shaped bistable region in the $(\mu,h)$ plane. Second, sit just past a saddle-node that has annihilated: the flow must squeeze through the corridor where the fixed points used to be, and the transit time diverges as the fold is approached — the ghost, quantified in the math and timed in Worked Example 2.

Figure

The imperfect pitchfork — a small symmetry-breaking h shatters the fork into a smooth branch plus a disconnected fold, bistable only inside the cusp wedge of the parameter plane

The core equations

The stage is a one-parameter flow on the line,

\dot{x}=f(x,\mu) \qquad (1)

where: $x$ — the reduced (center-manifold) coordinate near the criticality; $\mu$ — the control parameter, zeroed at the critical value. On the line, stability needs no matrix: a fixed point $x^*$ is stable when $f'(x^*)<0$ and unstable when $f'(x^*)>0$ — the slope of the crossing. Hyperbolic crossings cannot bifurcate. If $f'(x^*)\neq0$ , the implicit function theorem continues the root smoothly in $\mu$ with its slope's sign intact, so the portrait is locked. Every qualitative event therefore requires the double condition

f(x^*,\mu_c)=0 \quad\text{and}\quad \left.\frac{\partial f}{\partial x}\right|_{(x^*,\mu_c)}=0 \qquad (2)

where: the second equality is the tangency of the middle panel of the hero figure — the graph of $f$ kissing the axis. Two conditions, two unknowns $(x^*,\mu_c)$ : isolated degenerate points occur robustly as one knob turns, which is what "codimension one" means concretely. With no further structure, the generic event at such a point is the fold. Its normal form is

\dot{x}=\mu-x^2 \qquad (3)

where: for $\mu>0$ there are two fixed points $x^*=\pm\sqrt{\mu}$ with slopes $f'=\mp2\sqrt{\mu}$ — upper stable, lower unstable; at $\mu=0$ they fuse into a half-stable point; for $\mu<0$ the line is empty and the flow sweeps every initial condition to $-\infty$ . Derivation A proves universality: any smooth family satisfying (2) with generic quadratic terms rescales to (3) exactly. The fold's disappearance leaves measurable wreckage. Just past the annihilation ( $\dot{x}=\mu+x^2$ with small $\mu>0$ ), the transit time through the bottleneck where the equilibria died is

T=\int_{-\infty}^{\infty}\frac{\mathrm{d}x}{\mu+x^2}=\frac{\pi}{\sqrt{\mu}} \qquad (4)

where: the integral is dominated by the corridor $|x|\lesssim\sqrt{\mu}$ — the ghost of the vanished fixed points. The delay diverges as $\mu^{-1/2}$ : a system that has already tipped can linger near its dead state for arbitrarily long, the mechanism behind type-I intermittency and sluggish post-threshold transitions. Worked Example 2 puts numbers on it. If the physics pins an equilibrium at $x=0$ for every parameter value — $f(0,\mu)\equiv0$ identically — the constant term of the expansion is forbidden and the generic event becomes the exchange of stability:

\dot{x}=\mu x-x^2 \qquad (5)

where: the branches $x^*=0$ and $x^*=\mu$ have slopes $f'(0)=\mu$ and $f'(\mu)=-\mu$ : for $\mu<0$ the origin is stable and the other branch unstable, for $\mu>0$ the roles are exactly swapped. Nothing is created or destroyed — the branches pass through each other at the crossing, trading slopes. The laser threshold of the overview article and the epidemic threshold at basic reproduction number one are both this equation. Impose the stronger constraint of inversion symmetry, $f(-x,\mu)=-f(x,\mu)$ — a bead on a symmetric hoop, a column equally happy buckling left or right. Oddness kills the quadratic, and the leading nonlinearity is cubic:

\dot{x}=\mu x-x^3 \quad(\text{supercritical}), \qquad \dot{x}=\mu x+x^3 \quad(\text{subcritical}) \qquad (6)

where: the supercritical form sprouts the mirror pair $x^*=\pm\sqrt{\mu}$ for $\mu>0$ , born stable (Derivation C computes the slopes); the subcritical form instead has an unstable pair $\pm\sqrt{-\mu}$ for $\mu<0$ guarding a stable origin — and at $\mu=0$ the guard rails vanish and the state escapes to whatever distant physics saturates it. The minimal saturated subcritical model adds the first symmetric stabilizing term:

\dot{x}=\mu x+x^3-x^5 \qquad (7)

where: solving $\dot{x}=0$ for the nontrivial branches gives $x^{*2}=\tfrac12\bigl(1\pm\sqrt{1+4\mu}\bigr)$ — an S-shaped curve with folds at $\mu_s=-\tfrac14$ , $|x|=1/\sqrt{2}$ . Between $\mu_s$ and $0$ the system is bistable; sweeping the knob traces the hysteresis loop quantified in Derivation C. The jump at $\mu=0$ lands at $|x^*|=1$ , a finite amplitude — the signature of a hard transition. Finally, break the pitchfork's symmetry by an imperfection — a slight tilt of the hoop's axis, a load eccentricity, a stray field:

\dot{x}=h+\mu x-x^3 \qquad (8)

where: $h$ — the symmetry-breaking parameter. Fixed points solve $h=x^3-\mu x$ , a cubic whose number of real roots changes where double roots occur: at $3x^2=\mu$ , i.e. along the cusp

h^2=\frac{4\mu^3}{27} \qquad (9)

where: inside the cusp region ( $h^2<4\mu^3/27$ , requiring $\mu>0$ ) there are three equilibria and bistability; outside, exactly one. The perfect pitchfork is the single point at the cusp's tip approached along $h=0$ — which is why an unprotected pitchfork observed in a real experiment always resolves, under magnification, into a smooth branch passing near a fold.

Which normal form is it — the decision table

Structure of the problem	Forbidden terms	Normal form	Event
none (generic)	—	$\dot{x}=\mu-x^2$	fold: pair created/annihilated
equilibrium persists for all $\mu$	constant term	$\dot{x}=\mu x-x^2$	branches cross, swap stability
odd symmetry $f(-x)=-f(x)$	all even terms	$\dot{x}=\mu x\mp x^3$	mirror pair born (soft/hard)
odd symmetry, weakly broken	—	$\dot{x}=h+\mu x-x^3$	unfolded fork, cusp bistability

Derivation AEvery fold is the same fold — reduction to the saddle-node normal form

We prove the universality claim: any smooth one-parameter family with a nondegenerate tangency becomes $\dot{u}=\nu\pm u^2$ after smooth rescalings, so equation (3) is not an example but the general case. The strategy is Taylor expansion at the degenerate point followed by two changes of scale; the honest work is naming the nondegeneracy assumptions, because each one, when violated, spawns a different member of the trio. Watch how little survives: one sign.

Step 1. Put the degeneracy at the origin: $f(0,0)=0$ and $\partial_xf(0,0)=0$ by conditions (2). Expand to the lowest surviving orders: $f(x,\mu)=b\,\mu+a\,x^2+O(\mu x,\ x^3,\ \mu^2)$ , with $a=\tfrac12\partial_{xx}f(0,0)$ and $b=\partial_\mu f(0,0)$ . Why these two coefficients? The linear-in-x term is dead by hypothesis, so the quadratic $a$ is the leading restoring geometry, and $b$ measures how effectively the knob lifts the whole curve. Genericity is the double assumption $a\neq0$ , $b\neq0$ .

Step 2. Rescale the coordinate: with $u=|a|\,x$ , the flow becomes $\dot{u}=|a|\dot{x}=|a|b\,\mu+\operatorname{sgn}(a)\,u^2+\cdots$ . Define the new parameter $\nu=|a|\,b\,\mu$ : the family is now $\dot{u}=\nu+\operatorname{sgn}(a)\,u^2$ , and the residual sign flips under $u\to-u$ , $\nu\to-\nu$ , giving the canonical $\dot{u}=\nu-u^2$ .

Step 3. Confirm the neglected terms are harmless. Near the fold both $u$ and $\nu$ scale together — $u\sim\sqrt{\nu}$ on the branches — so a cubic correction contributes $u^3\sim\nu^{3/2}$ , subleading to both retained terms $\sim\nu$ . The count of fixed points, their $\pm\sqrt{\nu}$ positions to leading order, and the slopes $\mp2\sqrt{\nu}$ are untouched: the truncation is structurally stable as a one-parameter family.

Step 4. Read off what "generic" bought. Violate $b\neq0$ — the knob fails to lift the curve, as when an equilibrium is pinned for all $\mu$ — and the expansion starts $\mu x$ instead: the transcritical. Violate $a\neq0$ as well, by symmetry killing every even term, and the cubic leads: the pitchfork. The trio is not three unrelated phenomena but one Taylor expansion read under three different constraint regimes. Verified: matches the normal-form computation of Guckenheimer–Holmes §3.4 and Kuznetsov ch. 3; branch positions $u^*=\pm\sqrt{\nu}$ and slopes $\mp2\sqrt{\nu}$ recompute directly from the truncated form.

Go deeper · Derivation B · 2 minThe bead on the rotating hoop — a pitchfork you can build

Derivation BThe bead on the rotating hoop — a pitchfork you can build

The canonical laboratory pitchfork: a bead on a vertical circular hoop of radius $R$ , the hoop spun about its vertical diameter at rate $\omega$ , the bead's motion heavily damped (drag coefficient $b$ ). We derive the first-order flow for the bead's angle, expand it near the bottom, and watch normal form (6) emerge with computable coefficients — including the threshold $\omega_c=\sqrt{g/R}$ and the tilt angle of the buckled states. The sibling article Fixed Points and Stability in Phase Space treats this system as an energy landscape; here it becomes a worked bifurcation. The step to watch is the Taylor expansion: the symmetry does the censoring for us.

Step 1. In the frame rotating with the hoop, the bead at angle $\phi$ from the bottom feels gravity and the centrifugal effect, packaged in the effective potential $V_{\mathrm{eff}}(\phi)=mgR(1-\cos\phi)-\tfrac12mR^2\omega^2\sin^2\phi$ . In the strongly overdamped limit inertia is negligible and the flow is first-order: $b\,\dot{\phi}=-V_{\mathrm{eff}}'(\phi)=mR\sin\phi\,\bigl(R\omega^2\cos\phi-g\bigr)$ . Why overdamped? It reduces the dynamics to a flow on a line, the natural home of the 1-D normal forms; the equilibria and their count are the same with inertia, only the approach to them changes.

Step 2. Find all equilibria from the factored form: $\sin\phi=0$ gives the bottom $\phi=0$ and top $\phi=\pi$ for every $\omega$ ; the factor $R\omega^2\cos\phi=g$ gives the tilted pair $\phi^*=\pm\arccos\!\bigl(g/(R\omega^2)\bigr)$ , real only when $R\omega^2\geq g$ , i.e. $\omega\geq\omega_c=\sqrt{g/R}$ .

Step 3. Expand the flow about $\phi=0$ using $\sin\phi\approx\phi-\phi^3/6$ and $\cos\phi\approx1-\phi^2/2$ : $b\,\dot{\phi}=mR(R\omega^2-g)\,\phi-\tfrac12mR^2\omega^2\,\phi^3+O(\phi^5)$ , where the cubic coefficient keeps only its leading part near threshold. Note there is no $\phi^2$ term and there never could be: the hoop's mirror symmetry $\phi\to-\phi$ makes the flow odd.

Step 4. Match to normal form (6): $\dot{\phi}=\mu\phi-c\phi^3$ with $\mu=mR(R\omega^2-g)/b$ and $c=mR^2\omega^2/(2b)>0$ . The cubic is negative — supercritical. Crossing $\omega_c$ , the bottom loses stability ( $\mu$ changes sign) and the mirror pair is born at $\phi^{*2}=\mu/c=2(R\omega^2-g)/(R\omega^2)$ .

Step 5. Cross-check against the exact answer: expanding $\cos\phi^*=g/(R\omega^2)$ for small tilt gives $1-\phi^{*2}/2=g/(R\omega^2)$ , hence $\phi^{*2}=2(R\omega^2-g)/(R\omega^2)$ — identical to Step 4. The normal form is not an approximation to the threshold; it is the threshold, with the square-root growth law attached. Verified: threshold $\omega_c=\sqrt{g/R}$ and equilibrium $\cos\phi^*=g/(R\omega^2)$ reproduce Morin's rotating-hoop problem; the two routes to $\phi^{*2}$ agree at leading order.

Go deeper · Derivation C · 2 minSupercritical versus subcritical — the cubic sign, the jumps, and the hysteresis window

Derivation CSupercritical versus subcritical — the cubic sign, the jumps, and the hysteresis window

One sign in the normal form separates a gentle transition from a catastrophic one. We first classify the pitchfork by the cubic coefficient, then quantify the subcritical case with quintic saturation — locating the folds, the jump amplitudes, and the width of the bistable window that produces hysteresis. Everything is elementary algebra on polynomials; the physics payoff is a set of exact numbers an experimentalist can test.

Step 1. Take the odd flow $\dot{x}=\mu x+c\,x^3+O(x^5)$ . Nontrivial equilibria satisfy $x^{*2}=-\mu/c$ : they exist for $\mu>0$ when $c<0$ (supercritical) and for $\mu<0$ when $c>0$ (subcritical).

Step 2. Compute their stability once and for all: $f'(x^*)=\mu+3c\,x^{*2}=\mu+3c\,(-\mu/c)=-2\mu$ . For $c<0$ the branches live at $\mu>0$ , so $f'=-2\mu<0$ : stable. For $c>0$ they live at $\mu<0$ , so $f'=-2\mu>0$ : unstable. The formula $f'(x^*)=-2\mu$ is exact for the cubic form and worth memorizing. Why does the same expression decide both cases? Because the branches inherit their character from which side of the bifurcation they inhabit — the algebra automatically encodes "new states appear where they can be stable" only in the supercritical case.

Step 3. For the subcritical case add saturation, equation (7): $\dot{x}=\mu x+x^3-x^5$ . The outer equilibria solve $\mu+x^2-x^4=0$ , giving $x^{*2}=\tfrac12\bigl(1+\sqrt{1+4\mu}\bigr)$ (outer, large) and $x^{*2}=\tfrac12\bigl(1-\sqrt{1+4\mu}\bigr)$ (middle, existing only for $-\tfrac14\leq\mu\leq0$ ).

Step 4. Locate the folds where outer and middle branches merge: the square root vanishes at $\mu_s=-\tfrac14$ , where $x^{*2}=\tfrac12$ , i.e. $|x|=1/\sqrt{2}\approx0.707$ . Together with the pitchfork point at $\mu=0$ this brackets the bistable window $-\tfrac14<\mu<0$ in which the origin and the two outer states coexist.

Step 5. Trace the loop. Sweeping $\mu$ upward from below, the system rides $x=0$ until $\mu=0$ , where the origin destabilizes and the state jumps to the outer branch at $x^{*2}=\tfrac12(1+1)=1$ : a finite jump to $|x^*|=1$ . Sweeping back down, it rides the outer branch past $\mu=0$ (still stable there) until the fold at $\mu_s=-\tfrac14$ , then drops from $|x|=0.707$ back to the origin. Up-transition at $0$ , down-transition at $-\tfrac14$ : hysteresis of width $\tfrac14$ in $\mu$ , exactly the loop drawn in the overview article's figure. Verified: $f'(x^*)=-2\mu$ recomputes by substitution; the window edges $\mu_s=-1/4$ , jump values $|x|=1$ and $1/\sqrt{2}$ follow from the quadratic-in- $x^2$ equilibrium condition — Strogatz §3.4 quotes the same construction.

Limiting cases

Case	Result	Meaning
Saddle-node, $\mu>0$	$x^*=\pm\sqrt{\mu}$ , slopes $\mp2\sqrt{\mu}$	Pair of equilibria, one stable one not; slowing down as $\mu\to0^+$ .
Saddle-node vanished, $\mu>0$ in $\dot{x}=\mu+x^2$	transit time $T=\pi/\sqrt{\mu}$	The ghost bottleneck: divergent delay after the tipping point.
Transcritical at $\mu=0$	$\dot{x}=-x^2$	Half-stable origin; decay from $x>0$ is algebraic, $x\sim1/t$ , not exponential.
Pitchfork, supercritical	$x^*=\pm\sqrt{\mu}$ , $f'=-2\mu$	Soft symmetry breaking; amplitude grows continuously, no hysteresis.
Pitchfork, subcritical saturated	folds at $\mu=-\tfrac14$ , $\|x\|=\tfrac{1}{\sqrt2}$	Bistable window of width $\tfrac14$ ; jumps of finite amplitude; hysteresis.
Imperfect pitchfork, $h\neq0$	cusp $h^2=4\mu^3/27$	Fork unfolds into smooth branch + disconnected fold; three states only inside the cusp.
Large imperfection or $\mu<0$	single branch $x^*(\mu)$	The system tracks one equilibrium smoothly — no bifurcation visible at all.

Common trapThree branches in the diagram, so it must be a pitchfork

A three-pronged diagram proves nothing — the pitchfork is a symmetry theorem, not a picture. Unless the dynamics are exactly odd in $x$ (a physical mirror symmetry, not an approximate one), the generic three-branch structure is an imperfect fork: one branch that never bifurcates at all, plus a separate saddle-node pair, per equation (8). The difference is measurable, not cosmetic. A perfect fork forces the system to choose a side with amplitude growing as $\sqrt{\mu}$ from zero; an imperfect one steers it smoothly along the connected branch, and the "other" states are reachable only by a finite kick. Euler buckling looks like a textbook pitchfork, but any load eccentricity $h$ selects the bowing direction in advance — and near the cusp, equation (9) says the selection is decided by imperfections as small as $h\sim\mu^{3/2}$ , which for a 10% supercriticality means parts-per-thousand asymmetries. Always ask what enforces the symmetry before you draw the fork.

Worked examples

Example 1 — the rotating hoop, with numbers (confidence builder). A bead rides a hoop of radius $R=0.20$ m spinning about its vertical diameter. Find the critical spin rate, and the bead's equilibrium angle at $\omega=7.7\ \mathrm{rad\,s^{-1}}$ — by the normal form and exactly.

Step 1. Threshold: $\omega_c=\sqrt{g/R}=\sqrt{9.81/0.20}=\sqrt{49.05}=7.00\ \mathrm{rad\,s^{-1}}$ (about 67 rpm). Below this the bottom is the only stable equilibrium.

Step 2. At $\omega=7.7\ \mathrm{rad\,s^{-1}}$ : $R\omega^2=0.20\times59.29=11.86\ \mathrm{m\,s^{-2}}$ , comfortably above $g$ , so the bottom has destabilized and the mirror pair exists.

Step 3. Normal-form prediction (Derivation B, Step 4): $\phi^{*2}=2(R\omega^2-g)/(R\omega^2)=2(11.86-9.81)/11.86=2\times2.05/11.86=0.346$ , so $\phi^*=0.588\ \mathrm{rad}=33.7^\circ$ .

Step 4. Exact answer: $\cos\phi^*=g/(R\omega^2)=9.81/11.86=0.827$ , so $\phi^*=0.597\ \mathrm{rad}=34.2^\circ$ . The cubic truncation is off by $1.6\%$ at ten percent past threshold — the normal form is quantitative, not merely qualitative, well beyond the strict neighborhood of $\mu_c$ . $\omega_c=7.0\ \mathrm{rad\,s^{-1}}$ ; at $7.7\ \mathrm{rad\,s^{-1}}$ the bead sits at $\approx34^\circ$ , and the two answers agree to within 2%.

Example 2 — how long does a ghost delay the verdict? (exam level). A sensor's overdamped read-out variable obeys $\dot{x}=\mu+x^2$ (units: $x$ dimensionless, time in seconds), the aftermath of a saddle-node whose fixed points annihilated at $\mu=0$ . The device registers a "switch" when $x$ runs from large negative to large positive values. Compute the switching delay at $\mu=0.01$ and at $\mu=0.0001$ , and check the scaling law.

Step 1. Separate variables for the full passage: $T=\int_{-\infty}^{\infty}\mathrm{d}x/(\mu+x^2)=\bigl[\tfrac{1}{\sqrt{\mu}}\arctan(x/\sqrt{\mu})\bigr]_{-\infty}^{\infty}=\pi/\sqrt{\mu}$ — equation (4). The integral converges at both ends; virtually all of it accumulates in the corridor $|x|\lesssim\sqrt{\mu}$ where the ghost sits.

Step 2. At $\mu=0.01$ : $T=\pi/\sqrt{0.01}=\pi/0.1=31.4$ s.

Step 3. At $\mu=0.0001$ : $T=\pi/0.01=314$ s — over five minutes spent crawling through a bottleneck with no fixed point in it, while the dynamics far from the corridor take of order one second.

Step 4. Scaling check: reducing $\mu$ by a factor of 100 lengthened the delay by $\sqrt{100}=10$ , confirming $T\propto\mu^{-1/2}$ . In log–log form, delay data versus distance past the fold should fall on a line of slope $-\tfrac12$ — the standard experimental identification of a saddle-node ghost, used from semiconductor lasers (type-I intermittency) to conservation-biology models of delayed collapse. $T=31.4$ s at $\mu=10^{-2}$ , $314$ s at $\mu=10^{-4}$ : the inverse-square-root law in action.

Origins · 1 minLev Landau and how this was discovered

Lev Landau

1908–1968 · Kharkov and Moscow

Landau never used the word bifurcation, yet the supercritical pitchfork is his theory of second-order phase transitions in miniature. His 1937 papers proposed describing a transition by an order parameter $\psi$ and expanding the free energy as $F=a(T-T_c)\psi^2+b\psi^4$ — whose minima sit at $\psi=0$ above $T_c$ and at $\psi^*=\pm\sqrt{a(T_c-T)/2b}$ below: exactly the pitchfork diagram, with the mean-field exponent $\tfrac12$ this article derives as the square-root law. Symmetry dictating which terms may appear — the pitchfork's defining logic — is the organizing principle of the entire Landau school, and his classification of continuous transitions by broken symmetry groups is bifurcation theory's decision table written for $10^{23}$ particles. Superfluidity, superconductivity (through Ginzburg–Landau theory), liquid crystals, and the Higgs mechanism all wear this mathematics. Landau's career was itself discontinuous: imprisoned in 1938, released after Kapitza's intervention, Nobel laureate in 1962 for the theory of superfluid helium, his school's famous "theoretical minimum" exam shaping Soviet physics for a generation.

History

1744

— Euler computes the critical load of a compressed elastic column, the first pitchfork in the scientific record

1885

— Poincaré introduces the word bifurcation for the branching equilibrium figures of rotating fluid masses

1937

— Landau's theory of second-order phase transitions makes the supercritical pitchfork the template for symmetry breaking

1955

— Whitney proves the fold and the cusp are the only stable singularities of plane maps, grounding the imperfection theory

1972

— Thom's catastrophe theory popularizes the cusp geometry and its bistable interior

1976

— May's logistic-map review carries the same normal forms into ecology and the chaos era

Going further

References. Strogatz, Nonlinear Dynamics and Chaos (2nd ed.), Ch. 3 · Guckenheimer & Holmes, Nonlinear Oscillations, Dynamical Systems, and Bifurcations of Vector Fields, Ch. 3 · Wiggins, Introduction to Applied Nonlinear Dynamical Systems and Chaos, Ch. 20 · Morin, Introduction to Classical Mechanics (rotating-hoop problem) · Kuznetsov, Elements of Applied Bifurcation Theory, Ch. 3

Provenance. Normal forms and stability slopes ( $f'(x^*)=-2\mu$ for the cubic pitchfork; transcritical $f'(0)=\mu$ , $f'(\mu)=-\mu$ ) are exact and verifiable from the given $f$ . Rotating-hoop threshold $\omega_c=\sqrt{g/R}$ and equilibrium $\cos\phi^*=g/(R\omega^2)$ follow from $V_{\mathrm{eff}}'=0$ ; cross-check against Morin's rotating-hoop problem. Euler buckling load $P_c=\pi^2 EI/L^2$ is standard. Cusp boundary $h^2=4\mu^3/27$ comes from requiring a double root of $x^3-\mu x-h$ . A verifier should recompute the fixed-point sets and Jacobian signs for each normal form and confirm the catastrophe discriminant; cited: Strogatz Ch. 3, Guckenheimer & Holmes Ch. 3, Kuznetsov Ch. 3. Ghost passage time $T=\pi/\sqrt{\mu}$ is the standard $\int\mathrm{d}x/(\mu+x^2)$ result (Strogatz §4.3); worked numbers ( $\omega_c=7.00\ \mathrm{rad\,s^{-1}}$ at $R=0.20$ m; $T=31.4$ s at $\mu=0.01$ ) recompute directly.

Builds on: Fixed Points and Equilibria · Linear Stability Analysis and the Jacobian · Phase Portraits and the Phase Plane · Bifurcations and the Loss of Stability Leads to: Hopf Bifurcation and Birth of Limit Cycles · Period-Doubling Cascade and the Route to Chaos · Feigenbaum Constants and Universality · Limit Cycles and Self-Sustained Oscillations

Why is it called a saddle-node when a one-dimensional flow has no saddles?

The name is inherited from the smallest phase space where the event shows its full anatomy: the plane. There the colliding pair generically consists of a saddle and a node — trajectories funnel into the node while the saddle's stable manifold walls off the basin, and at the collision the two fuse into a saddle-node point before vanishing. The one-dimensional form $\dot{x}=\mu-x^2$ is the dynamics along the center direction of that planar event; the node direction has been slaved away by the center-manifold reduction of the overview article. Some books say "fold" or "tangent bifurcation" precisely to avoid the mismatch; all three names label the same normal form.

The transcritical seems to depend on my model being exactly right — is it physically robust?

It is robust exactly as often as the pinning constraint is a law rather than an approximation. "Zero photons emit nothing," "zero infected infect nobody" — these are structural facts, and systems governed by them genuinely exhibit transcritical exchanges. The caveat is that many real systems only approximately pin the state: spontaneous emission, imported infections, stray fields add a small constant $\varepsilon$ to $f$ , and the crossing unfolds into two hyperbolae that avoid each other — the branches swap identities smoothly. Observationally you still see a sharp threshold with critical slowing down; you just never see the exact kink. The modeling advice: state which conservation or emptiness principle pins your branch, and estimate $\varepsilon$ before trusting the perfect diagram.

What actually happens exactly at the critical parameter, where the linear term is gone?

Each normal form answers differently, and the difference is measurable. At the saddle-node tangency the merged point is half-stable: perturb right, you relax back; perturb left, you escape. At the transcritical and supercritical pitchfork points the origin is still stable, but nonlinearly so — $\dot{x}=-x^2$ (one-sided) or $\dot{x}=-x^3$ — and relaxation is algebraic, $x\sim1/t$ or $x\sim t^{-1/2}$ , instead of exponential. That loss of an exponential rate is critical slowing down in its purest form: the linear clock has stopped, and only nonlinear friction remains. At the subcritical point the origin is nonlinearly unstable ( $\dot{x}=+x^3$ ), which is why the transition there is violent. Same eigenvalue (zero) in all four cases; four different fates — the definitive proof that the linearization cannot adjudicate criticality.

Is the pitchfork really the same mathematics as Landau's phase transitions?

Structurally yes; the dictionary is worth writing once: order parameter $\psi\leftrightarrow x$ , reduced temperature $(T_c-T)\leftrightarrow\mu$ , and free-energy relaxation $\dot{\psi}=-\partial F/\partial\psi$ with $F=a(T-T_c)\psi^2+b\psi^4$ giving exactly $\dot{x}=\mu x-x^3$ up to scales. Supercritical = second-order transition (continuous $\psi$ , exponent $\tfrac12$ ); subcritical with saturation = first-order (jump, metastability, hysteresis); the imperfection $h$ = an external field, and the cusp is the coexistence wedge. The physical difference is fluctuations: Landau's version is mean-field for $10^{23}$ degrees of freedom, and near $T_c$ real critical exponents deviate from $\tfrac12$ , whereas for a low-dimensional dynamical system the square root is exact — why the hoop matches theory to percent accuracy while a ferromagnet does not.

Quantitatively, how small an imperfection destroys the pitchfork?

Smaller than intuition suggests, and the cusp gives the exact budget: the three-state region survives only while $|h|<2\mu^{3/2}/(3\sqrt{3})\approx0.385\,\mu^{3/2}$ . Ten percent past threshold ( $\mu=0.1$ scaled) tolerates $|h|\lesssim0.012$ ; at $\mu=0.01$ the budget is $4\times10^{-4}$ . Because the tolerance shrinks as the three-halves power, every physical fork looks perfect far beyond threshold and imperfect near it: sweep slowly through $\mu_c$ and the system does not choose a side at random — it follows the branch pre-selected by $h$ , however tiny. Experiments wanting genuine spontaneous symmetry breaking must cross the threshold fast (so fluctuations beat the bias) or shield the symmetry to extraordinary tolerance — exactly the protocol in spinor-condensate and buckling experiments.

Why should I care about the ghost if the fixed points no longer exist?

Because the ghost sets the time budget of every process that has just tipped. The corridor where the equilibria died still has $|\dot{x}|\approx\mu$ — arbitrarily slow — so the system behaves for a time $\pi/\sqrt{\mu}$ as if the dead state were alive. Consequences: a laser just above the intermittency threshold emits long laminar stretches interrupted by bursts, with mean laminar length obeying the $\mu^{-1/2}$ law (the standard signature of type-I intermittency); ecosystems past a tipping point can look healthy for decades — extinction debt — because they are transiting the ghost; and control systems that must react to a tip get a computable grace period. Measure a delay scaling as the inverse square root of the distance past threshold and you have diagnosed a saddle-node without ever seeing the fixed points.

Do these one-dimensional results actually govern my high-dimensional system?

Yes, near the bifurcation — that is the center-manifold theorem doing its work, as set out in the overview article. When one real eigenvalue crosses zero and the other $n-1$ stay strictly negative, the fast directions are exponentially slaved and the whole local problem reduces to a one-dimensional flow along the slow curve, where this article's trio is the complete generic list. Two limitations: the reduction is local (the distant attractor a subcritical jump lands on lives outside it), and it needs the spectral gap (a second near-zero eigenvalue puts you at a codimension-two point — cusp, Takens–Bogdanov — with richer normal forms). Within those bounds, buckling shells, power grids, and neuron models all reduce, near their static thresholds, to $\dot{x}=\mu-x^2$ or its two constrained cousins.

Can hysteresis ever occur with a supercritical pitchfork?

Not from the pitchfork itself — that is the operational meaning of "soft." Above threshold the only stable states are the branches at $\pm\sqrt{\mu}$ , continuously connected to the symmetric state; sweep $\mu$ up and down and the amplitude retraces the same square-root curve, because at every $\mu$ there is a unique available amplitude (up to sign). Hysteresis needs coexisting attractors at fixed $\mu$ , which requires a fold — supplied by a subcritical cubic with quintic saturation (window $-\tfrac14<\mu<0$ , Derivation C) or by an imperfection tilting the diagram inside the cusp. One caveat: sweeping fast produces dynamic lag that mimics hysteresis even supercritically — the state needs time $\sim1/|f'|$ to follow the branch, and near threshold that diverges. Distinguish by sweeping slower: real hysteresis converges to a finite loop, lag shrinks toward zero.

Saddle-Node, Transcritical, and Pitchfork Bifurcations

Where this lives today

The physical picture

The core equations

Which normal form is it — the decision table

Limiting cases

Worked examples

Going further

Before this

Where this leads