I’ve been thinking a lot about how I approach problems lately, and I realized that I often fall into the trap of only thinking in one mode. I’m either in “divergent mode,” where I’m just generating ideas and not really evaluating them, or I’m in “convergent mode,” where I’m trying to find the best solution and not really generating new ideas. I’ve only recently come to the realization that it’s important to be able to switch between the two modes, and I’m trying to get better at it.

Research is not a linear process. I used to think of it as a linear process, but it’s really just a dance between two fundamentally different modes of thinking: divergent thinking and convergent thinking. Understanding when to use each and how to move fluidly between them is perhaps the most important meta-skill for any researcher, engineer, or creative professional.

Think of these modes as two sides of the same coin or 2 ends of a spectrum. Too much divergent thinking, and you chase every interesting tangent without ever completing anything. Too much convergent thinking, and you optimize yourself into a local maximum, missing breakthrough ideas that require you to question your assumptions.

In this article i’ll share with you how i’ve come to understand these two modes, their cognitive underpinnings, and practical strategies for balancing them in problem-solving and research in general. I’ll also share some resources that have helped me along the way.

What Is Divergent Thinking?

Divergent thinking it is simply put, the cognitive process of generating multiple, novel solutions to open-ended problems. It’s usually characterized by:

  • Ideation without judgment: where you defer evaluating an idea to maximize idea generation
  • Exploration: Investigating multiple directions simultaneously
  • Tolerance for ambiguity: where you’re comfortable with uncertainty and incomplete information
  • Questioning assumptions: where you challenge existing frameworks and constraints

J.P. Guilford first introduced divergent thinking in 1950 as part of his Structure of Intellect theory, distinguishing it from intelligence measured by traditional IQ tests [1]. Unlike convergent thinking (which seeks the correct answer), divergent thinking asks: What are all the possible answers?

Examples in Research

In research contexts, divergent thinking manifests as:

  • What if we represented state differently?
    Instead of using environment models in operational semantics, what if we used substitution? What if we used graph-based representations?

  • What happens if we remove this assumption?
    What if we don’t assume call-by-value? What if types don’t have to be decidable? What if we allow effects in pure languages?

  • How would category theory view this problem?
    Can we model this computation as a functor? Is there a monad here? What about coalgebras?

This mode is essential during:

  • Early-stage research when exploring a problem space
  • Brainstorming sessions
  • Literature review (discovering connections across fields)
  • Designing experiments with multiple hypotheses

Eventually, i figured out my default thinking mode is divergent thinking. I’m always generating ideas and not really evaluating them. I’m always exploring multiple directions simultaneously. I’m always comfortable with uncertainty and incomplete information. I’m always questioning assumptions. This is great for research, but it’s not so great for writing papers or giving presentations or even focusing on the work i’m paid to do everyday. I find myself starting a task and then thinking about 150 million ways to complete it and end not not even writing a single line of code or a single word of the paper.

What Is Convergent Thinking?

Convergent thinking is the cognitive process of finding the single best solution to a well-defined problem. It’s characterized by:

  • Critical evaluation: Analyzing ideas for validity and feasibility
  • Deductive reasoning: Applying logic to narrow down possibilities
  • Optimization: Refining solutions to meet specific criteria
  • Verification: Testing whether ideas actually work

Guilford contrasted divergent thinking with convergent thinking, noting that conventional education and testing primarily measure the latter [1]. Convergent thinking asks: Which answer is correct?

Examples in Research

We can see convergent thinking in action when:

  • Does this actually work?
    Can I implement this formalism? Does my type system ensure soundness? Can this algorithm handle edge cases?

  • Can I prove this property?
    Is progress preserved? Does type safety hold? Can I show termination?

  • Is this simpler than existing approaches?
    Does my abstraction reduce cognitive overhead? Is the implementation more maintainable?

This mode is essential during:

  • Proof development
  • Implementation and debugging
  • Writing formal arguments
  • Evaluating experimental results
  • Preparing submissions (when clarity and rigor are paramount)

The Cognitive Science Behind the Modes

Modern neuroscience reveals that these two modes engage different neural networks:

Divergent Thinking: The Default Mode Network

Divergent thinking heavily relies on the Default Mode Network (DMN)—a set of brain regions active during rest, daydreaming, and spontaneous thought [2]. The DMN is associated with:

  • Autobiographical memory
  • Future planning and simulation
  • Mind-wandering
  • Semantic processing

Research using fMRI has shown increased DMN activation during creative ideation tasks, particularly in the medial prefrontal cortex and posterior cingulate cortex [3]. This explains why breakthrough ideas often come during walks, showers, or other “unfocused” activities.

Convergent Thinking: The Executive Control Network

Convergent thinking engages the Executive Control Network (ECN), particularly the dorsolateral prefrontal cortex, which is responsible for:

  • Working memory
  • Attention control
  • Goal-directed behavior
  • Inhibition of irrelevant information

Studies show that the ECN activates during tasks requiring focused problem-solving, logical reasoning, and decision-making [4]. This is the network active when you’re debugging code, writing proofs, or evaluating competing hypotheses.

The Dynamic Interplay

Crucially, creative insight requires both networks working in coordination. Beaty et al. (2016) found that highly creative individuals show stronger functional connectivity between the DMN and ECN [5]. This suggests that creativity isn’t just about free association but more about strategically coupling exploration with evaluation. This is what i’m trying to do with my research and with the advent of AI and coding agents, i’m usually able to quickly prototype and test the convergent thinking part of my research.

The Research Lifecycle: When to Use Each Mode

Research requires both modes, but at different stages:

Phase 1: Problem Discovery (Divergent)

When i’m starting a research project, divergent thinking helps me:

  • Identify interesting questions
  • Survey the landscape of existing work
  • Generate hypotheses
  • Discover unexpected connections

Strategy:

  • Read broadly across fields
  • Maintain a “curiosity list” of open questions
  • Engage in discussions without immediate goal
  • Allow time for unstructured exploration

Phase 2: Problem Formulation (Convergent → Divergent)

Once i’ve identified a problem, i narrow it down to a set of questions:

  • What specific question am I answering?
  • What are the constraints and assumptions?
  • What would constitute a solution?

Then i diverge again:

  • What are multiple potential approaches?
  • What related problems might inform this one?

Strategy:

  • Write a one-paragraph problem statement
  • List 10+ different approaches (try to force the divergence)
  • Map connections to other domains (usually things i’m already working on, read about or worked in the past)

Phase 3: Solution Development (Oscillating)

Implementation requires me to rapidly cycle between divergent and convergent thinking:

  • Diverge: Try different architectural approaches
  • Converge: Debug and prove correctness
  • Diverge: Consider alternative optimizations
  • Converge: Benchmark and validate

Strategy:

  • Time-box divergent exploration (e.g., 2 hours), this is usually the most difficult part of me because my brain just get’s stuck and overthinking and i end up not getting anything done.
  • I switch to convergent implementation and try to implement the simplest approach i came up with.
  • Then i Iterate based on what i learn

Phase 4: Evaluation and Refinement (Convergent)

When writing or building production systems:

  • I rigorously test my claims
  • Prove properties by hand or with a tool
  • then compare to baselines
  • then simplify and clarify

Strategy:

  • I adopt a “devil’s advocate” mentality
  • then i seek counterexamples actively
  • then i refactor for clarity and elegance

Phase 5: Communication (Convergent with Strategic Divergent Touches)

Papers require clarity (convergent), but the best papers also:

  • Motivate with surprising examples (divergent framing)
  • Connect to unexpected domains (divergent insight)
  • Suggest future work (divergent possibilities)

Practical Techniques for Balancing the Modes

These are the techniques i use to balance the modes.

For Divergent Thinking

  1. The 10x10 Method
    Generate 10 ideas in 20-30 minutes. Force myself beyond the obvious first 3-4 ideas.

  2. Constraint Removal
    Ask myself: “What if time/money/compute were unlimited?” Then i try to work backwards.

  3. Cross-Domain Analogies
    How would a biologist/economist/artist approach this problem?

  4. The “Yes, and…” Rule
    In brainstorming, i try to build on ideas rather than critiquing them immediately.

  5. Scheduled Unstructured Time
    I block 2-3 hours per day for “exploration without agenda.”

For Convergent Thinking

  1. The Feynman Technique
    Explain my idea simply. Where i struggle indicates gaps in understanding.

  2. Proof by Counterexample
    Actively try to break my own ideas. If i can’t, they’re stronger (at least i know i’m not wrong “Yet”).

  3. The Five Whys
    Ask “why” repeatedly to get to root causes and validate assumptions.

  4. Forced Simplification
    Can i cut 50% of my code/proof/argument? What’s the minimal version?

  5. Peer Review Simulation
    Anticipate every objection a reviewer might raise.

Common Pathologies and How to Recognize Them

Too Much Divergence: “The Endless Exploration”

Symptoms:

  • Constantly starting new projects without finishing
  • Inability to commit to a direction
  • Analysis paralysis from too many options
  • Papers perpetually in “draft” status

Remedy:

  • Set explicit decision deadlines
  • Use commitment devices (announce plans publicly)
  • Force yourself to write a complete proof or implementation
  • Timebox exploration phases

Too Much Convergence: “The Local Maximum Trap”

Symptoms:

  • Incremental improvements without breakthroughs
  • Reluctance to question foundational assumptions
  • Optimizing solutions that solve the wrong problem
  • Dismissing ideas too quickly

Remedy:

  • Schedule monthly “assumption audits”
  • Force yourself to generate 5 alternative approaches before committing
  • Read outside your subfield regularly
  • Collaborate with people from different backgrounds

Case Study: Type Systems Research

Let me illustrate with a concrete example from programming language theory.

Divergent Phase: Early type system designers asked expansive questions:

  • What properties should types guarantee? (Curry-Howard correspondence)
  • What if types could express more than just “int” or “string”? (Dependent types)
  • How do types relate to logic? (Linear logic → linear types)
  • What if types could track effects? (Effect systems)

Convergent Phase: Each idea required rigorous development:

  • Prove type soundness (“well-typed programs don’t go wrong” [6])
  • Formalize semantics
  • Build proof assistants to verify the theory
  • Implement practical type checkers

Oscillation: Modern research cycles between modes:

  • Divergent: Gradual typing, refinement types, session types, algebraic effects
  • Convergent: Each requires soundness proofs, decidability results, implementation
  • Divergent: What’s the unifying framework? Category theory? Game semantics?
  • Convergent: Make it practical. How do we infer types? Optimize checking?

Robin Milner’s development of ML demonstrates this beautifully [7]. The initial divergent insight (“LCF needs a meta-language with strong guarantees”) led to convergent work (Hindley-Milner type inference), which enabled new divergent exploration (polymorphism as a foundation for modularity).

Cultivating Both Modes

Research/Creative excellence requires strength in both modes:

For Naturally Divergent Thinkers

If you’re idea-rich but execution-poor:

  • Set clear scopes: Define “done” before starting
  • Embrace constraints: Artificial limits force completion
  • Pair with convergent thinkers: Collaborate with detail-oriented colleagues, i.e. accountability partners
  • Track completion rate: Measure finished vs. started projects

For Naturally Convergent Thinkers

If you’re rigorous but struggle with novelty:

  • Schedule exploration time: Treat it as mandatory, not optional
  • Lower the stakes: Do side projects with no publication pressure
  • Change environments: Work in cafes, parks, or different offices
  • Read fiction/poetry: Train pattern-matching in different domains

The Meta-Skill: Recognizing Which Mode You Need

The hallmark of expertise is knowing when to use each mode.

Use Divergent Thinking When:

  • You’re stuck in a rut
  • All approaches seem equally mediocre
  • The problem feels ill-defined
  • You’re starting a new project
  • You need motivation/excitement

Use Convergent Thinking When:

  • You have too many half-finished ideas
  • Claims need rigorous validation
  • You’re preparing to ship/publish
  • Debugging or proving correctness
  • You need to simplify and clarify

Switch Modes When:

  • You’ve been in one mode for >3 hours
  • You feel frustrated or stuck
  • You’ve exhausted current approach
  • You notice diminishing returns

Conclusion: The Dance of Research

Research is not about choosing between creativity and rigor—it’s about dancing between them. The most impactful work comes from researchers who can:

  1. Diverge boldly: Explore weird ideas without immediate judgment
  2. Converge ruthlessly: Validate relentlessly and simplify aggressively
  3. Transition smoothly: Recognize when to switch modes
  4. Cycle rapidly: Move fluidly between exploration and exploitation

Too much divergence leads nowhere. Too much convergence kills creativity. But the right balance, the strategic oscillation between opening up possibilities and narrowing them down—is where breakthroughs happen.

The next time you sit down to work, ask yourself: Which mode does this problem need right now?

Then give yourself permission to fully inhabit that mode, knowing you’ll balance it with the other when the time is right.

References

[1] Guilford, J. P. (1950). Creativity. American Psychologist, 5(9), 444–454. https://doi.org/10.1037/h0063487

[2] Raichle, M. E., et al. (2001). A default mode of brain function. Proceedings of the National Academy of Sciences, 98(2), 676-682. https://doi.org/10.1073/pnas.98.2.676

[3] Beaty, R. E., et al. (2014). Creativity and the default network: A functional connectivity analysis of the creative brain at rest. Neuropsychologia, 64, 92-98. https://doi.org/10.1016/j.neuropsychologia.2014.09.019

[4] Cole, M. W., & Schneider, W. (2007). The cognitive control network: Integrated cortical regions with dissociable functions. NeuroImage, 37(1), 343-360. https://doi.org/10.1016/j.neuroimage.2007.03.071

[5] Beaty, R. E., et al. (2016). Robust prediction of individual creative ability from brain functional connectivity. Proceedings of the National Academy of Sciences, 113(4), 1087-1092. https://doi.org/10.1073/pnas.1713532114

[6] Milner, R. (1978). A theory of type polymorphism in programming. Journal of Computer Science and System Science, 17, 348-375.

[7] Milner, R., Tofte, M., Harper, R., & MacQueen, D. (1997). The Definition of Standard ML (Revised). MIT Press.

Further Reading:

  • Sawyer, R. K. (2011). Explaining Creativity: The Science of Human Innovation (2nd ed.). Oxford University Press.
  • Kaufman, S. B., & Gregoire, C. (2015). Wired to Create: Unraveling the Mysteries of the Creative Mind. TarcherPerigee.
  • Csikszentmihalyi, M. (1996). Creativity: Flow and the Psychology of Discovery and Invention. Harper Collins.
  • Pierce, B. C. (2002). Types and Programming Languages. MIT Press. (For the type systems case study)