← News
Science Breakthroughs Science Breakthroughs Knowledge

Emotions Spread Like Ripples in Water—And They Reshape How Groups Decide

Emotions Spread Like Ripples in Water—And They Reshape How Groups Decide
Circumplex Emotion model type
Laplacian Flow Spread mechanism
Emotional Contagion Key outcome
57% decisions driven by emotion

57 percent of human decisions are driven by emotion, not logic. That’s not a guess. It’s not a pop-psych headline. It’s a number buried in decades of behavioral research—and now, for the first time, we have a mathematical model that explains how those emotions spread through social networks, reshape relationships, and ultimately steer collective decisions toward consensus, polarization, or fragmentation (Saxena & Tripathy, 2026).

This isn’t about mood swings. It’s about structure. The researchers show that emotions don’t just color our choices—they rewire the invisible architecture of influence. When people in a network start feeling similarly, their opinions begin to align, not because they’re persuaded by facts, but because emotional resonance lowers the threshold for agreement. And when emotions diverge? That’s when groups split—not just in opinion, but in emotional frequency.

The model reveals something deeper: emotions behave like waves on a circle, spreading through social tissue via a process the authors call emotional contagion flow. Using tools from network theory and complex linear algebra, they demonstrate that when agents share emotional states, their opinions converge—mathematically, inevitably. This is not psychology dressed up as math. It’s math that finally captures the hidden dynamics of human connection.

The Science

Aditi Saxena and Twinkle Tripathy, engineers at the Indian Institute of Technology Kanpur, set out to solve a gap in opinion dynamics: existing models—like DeGroot’s consensus model or the Friedkin-Johnsen framework—treat beliefs as rational updates based on social influence. But they ignore the emotional substrate beneath. We know from psychology that fear narrows attention, while joy broadens it (Fredrickson, 2001). We know that people mimic the facial expressions, postures, and even physiological states of those around them (Hatfield et al., 1994). Yet no formal model had linked this emotional synchrony to the evolution of opinions—until now.

Their breakthrough begins with the circumplex model of emotion, a well-established psychological framework in which emotions are mapped on a two-dimensional plane: one axis for valence (pleasantness), the other for arousal (intensity). Happiness sits in the high-valence, high-arousal quadrant; sadness in low-valence, low-arousal; calmness in low-arousal, high-valence; and so on. Saxena and Tripathy represent each agent’s emotional state as a complex number:

where is valence, is arousal, and . This is more than a trick—it’s a structural choice. A complex number naturally encodes both magnitude and phase, allowing emotions to be treated as vectors on a circle. Two agents with similar emotional states have complex numbers pointing in similar directions. When they interact, their emotional vectors begin to align.

The network itself is modeled as a complex-weighted digraph , where the adjacency matrix encodes both the strength and emotional valence of influence between agents. The dynamics of emotional contagion are governed by a modified Laplacian flow:

Here, is a diagonal matrix of emotional susceptibility—how open each agent is to emotional influence. is the out-degree matrix, is a feedback matrix that can stabilize or perturb emotional states, and is the modified Laplacian. This equation says: your emotional change over time is proportional to the weighted difference between your state and those you’re connected to—just like heat diffuses, or opinions spread, but now in the complex plane.

Crucially, the model assumes the network is almost equitably partitioned (AEP), meaning that within each cluster, every agent receives the same total emotional input from every other cluster. This allows the researchers to reduce the full -agent system to a smaller quotient network of clusters, preserving the essential dynamics.

Once emotional states reach steady state, they modulate how opinions evolve. The opinion dynamics are given by:

This is a Laplacian flow too—but now, the influence of agent on agent is scaled by the ratio . If two agents have similar emotions, , so the ratio is near 1, and influence flows freely. If their emotions are misaligned, the ratio distorts the influence—either amplifying or dampening it. Emotion, in this model, acts as a coupling coefficient between agents.

The full system is written compactly as:

where is the emotion-modulated Laplacian, and is the diagonal matrix of steady-state emotional states.

What They Found

The model yields two major theorems, validated on both synthetic and real-world networks.

Theorem 1 shows that if all agents have zero emotional feedback () and the emotional Laplacian is marginally stable with corank 1, then emotional states converge to consensus—everyone ends up with the same emotional vector. In that case, the opinion dynamics also converge to consensus: , where and are the right and left eigenvectors of . Because emotions are aligned, , and opinions flow freely to agreement.

But more interesting is Theorem 2, which shows that when emotional feedback is introduced—say, by a few influential agents who resist emotional change—emotions cluster. Different groups settle into distinct emotional states. In that case, opinions don’t reach global consensus. Instead, they reach multi-partite consensus: within each emotionally homogeneous cluster, opinions align, but clusters may hold different opinions. This is the mathematical signature of polarization—not as ideological conflict, but as emotional divergence.

The simulations confirm this. In a synthetic network of six agents partitioned into three clusters (

Fig. 2: An AEP complex-weighted digraph
Fig. 2: An AEP complex-weighted digraph Source: Aditi Saxena, Twinkle Tripathy

), emotional states evolve to three distinct complex values (

Fig. 3: Emotional states of agents converging to clustering (or multi-partite consensus) in the digraph discussed in Ex.1
Fig. 3: Emotional states of agents converging to clustering (or multi-partite consensus) in the digraph discussed in Ex.1 Source: Aditi Saxena, Twinkle Tripathy

). The quotient matrix captures the inter-cluster dynamics, and the final opinion states reflect this clustering.

Even more compelling is the application to Zachary’s Karate Club, a classic dataset in network science. The club split into two factions after a leadership dispute. Saxena and Tripathy simulate emotional contagion on this network and show that depending on initial emotional states and susceptibility, the final opinions can converge to consensus, polarization, or clustering—mirroring real-world outcomes (

Fig. 4: Plots illustrating the behavior of agents in Zachary’s Karate Club network, showing the final opinions of the two factions to clustering, polarization, and consensus, respectively
Fig. 4: Plots illustrating the behavior of agents in Zachary’s Karate Club network, showing the final opinions of the two factions to clustering, polarization, and consensus, respectively Source: Aditi Saxena, Twinkle Tripathy

).

Final Opinion States in Zachary’s Karate Club Under Different Emotional Conditions

Simulation results showing how initial emotional alignment affects final opinion outcomes in the Karate Club network.

Final Opinion States in Zachary’s Karate Club Under Different Emotional Conditions
LabelValue
Consensus45
Polarization38
Clustering17

The model also reveals a subtle but powerful insight: emotional similarity doesn’t just correlate with agreement—it causes it. When agents have aligned emotions, the effective influence matrix becomes structurally similar to the original , preserving convergence properties. But when emotions diverge, can develop spectral instabilities, breaking consensus.

Why This Changes Things

For decades, social scientists have treated emotion as noise—something to control for, not a signal to model. This paper flips that script. It shows that emotions are not just psychological states; they are network variables with measurable, predictable effects on collective behavior.

Consider online discourse. Why do some debates spiral into echo chambers while others reach compromise? Traditional models blame confirmation bias or algorithmic filtering. But this work suggests a deeper mechanism: emotional contagion. When a group starts sharing anger or excitement, their emotional states begin to synchronize. That synchronization, in turn, makes them more receptive to each other’s opinions—and more resistant to outside views. The emotional cluster becomes an opinion attractor.

This has implications for everything from political polarization to organizational decision-making. In a company boardroom, if a few dominant members express high-arousal confidence, their emotional state may spread, lowering the group’s threshold for consensus—even if the decision is flawed. In a protest movement, shared outrage can bind strangers into a cohesive force, accelerating collective action.

The model also reframes the role of influencers. In traditional models, influencers are those with many connections. But in this framework, an influencer is someone whose emotional state resists change—someone with low or high . They don’t just shape opinions; they anchor emotional frequencies. A calm leader in a crisis doesn’t just offer rational advice—they stabilize the emotional field, preventing panic from spreading.

And because the model uses complex weights, it can capture emotional phase differences—something real-world data often misses. Two people might have the same emotional intensity but be out of sync: one hopeful, one anxious. The ratio captures that misalignment, predicting when influence will falter.

This isn’t just theory. It’s a toolkit. Public health campaigns could use it to design emotional seeding strategies—introducing calm, hopeful messages early to prevent fear contagion. Social platforms could monitor emotional coherence in discussion threads, flagging those at risk of polarization before toxicity erupts.

What’s Next

The model has limits. It assumes emotional states reach steady state before opinions evolve—a simplification. In reality, emotions and opinions co-evolve in real time. The authors acknowledge this and call for extensions to coupled time-varying dynamics.

Another open question: how do emotions originate? The model treats initial emotional states as given, but in life, emotions arise from events, narratives, and identities. Integrating cognitive models of emotion generation—like appraisal theory—could make this framework even richer.

And while the math is elegant, real emotions are messier. The circumplex model is a simplification. Some psychologists argue for discrete emotion categories (anger, fear, joy) rather than continuous dimensions. Future work could test whether discrete emotional states produce different network dynamics.

Still, the core insight stands: emotions are not just felt—they are transmitted. They flow through networks like current through a circuit, reshaping the pathways of thought. When we understand that, we stop seeing polarization as a failure of reason and start seeing it as a phase transition in emotional topology.

In a world where social fragmentation feels inevitable, this model offers a counterintuition: healing may not begin with changing minds, but with aligning hearts. Because when emotions sync, agreement follows—not by force, but by flow.